R for Beginners

advertisement
R for Beginners
Emmanuel Paradis
Institut des Sciences de l’Évolution
Université Montpellier II
F-34095 Montpellier cédex 05
France
E-mail: paradis@isem.univ-montp2.fr
I thank Julien Claude, Christophe Declercq, Élodie Gazave, Friedrich Leisch,
Louis Luangkesron, François Pinard, and Mathieu Ros for their comments and
suggestions on earlier versions of this document. I am also grateful to all the
members of the R Development Core Team for their considerable efforts in
developing R and animating the discussion list ‘rhelp’. Thanks also to the
R users whose questions or comments helped me to write “R for Beginners”.
Special thanks to Jorge Ahumada for the Spanish translation.
c 2002, 2005, Emmanuel Paradis (12th September 2005)
Permission is granted to make and distribute copies, either in part or in
full and in any language, of this document on any support provided the above
copyright notice is included in all copies. Permission is granted to translate
this document, either in part or in full, in any language provided the above
copyright notice is included.
Contents
1 Preamble
1
2 A few concepts before starting
2.1 How R works . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Creating, listing and deleting the objects in memory . . . . . .
2.3 The on-line help . . . . . . . . . . . . . . . . . . . . . . . . . .
3
3
5
7
3 Data with R
3.1 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Reading data in a file . . . . . . . . . . . . . . . . . . . . . .
3.3 Saving data . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Generating data . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1 Regular sequences . . . . . . . . . . . . . . . . . . . .
3.4.2 Random sequences . . . . . . . . . . . . . . . . . . . .
3.5 Manipulating objects . . . . . . . . . . . . . . . . . . . . . . .
3.5.1 Creating objects . . . . . . . . . . . . . . . . . . . . .
3.5.2 Converting objects . . . . . . . . . . . . . . . . . . . .
3.5.3 Operators . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.4 Accessing the values of an object: the indexing system
3.5.5 Accessing the values of an object with names . . . . .
3.5.6 The data editor . . . . . . . . . . . . . . . . . . . . . .
3.5.7 Arithmetics and simple functions . . . . . . . . . . . .
3.5.8 Matrix computation . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
11
14
15
15
17
18
18
23
25
26
29
31
31
33
4 Graphics with R
4.1 Managing graphics . . . . . . . . . . . .
4.1.1 Opening several graphical devices
4.1.2 Partitioning a graphic . . . . . .
4.2 Graphical functions . . . . . . . . . . . .
4.3 Low-level plotting commands . . . . . .
4.4 Graphical parameters . . . . . . . . . .
4.5 A practical example . . . . . . . . . . .
4.6 The grid and lattice packages . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
36
36
36
37
40
41
43
44
48
5 Statistical analyses with R
5.1 A simple example of analysis of variance
5.2 Formulae . . . . . . . . . . . . . . . . .
5.3 Generic functions . . . . . . . . . . . . .
5.4 Packages . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
55
56
58
61
6 Programming with R in pratice
6.1 Loops and vectorization . . . . . . . . . . . . . . . . . . . . . .
6.2 Writing a program in R . . . . . . . . . . . . . . . . . . . . . .
6.3 Writing your own functions . . . . . . . . . . . . . . . . . . . .
64
64
66
67
7 Literature on R
71
1
Preamble
The goal of the present document is to give a starting point for people newly
interested in R. I chose to emphasize on the understanding of how R works,
with the aim of a beginner, rather than expert, use. Given that the possibilities
offered by R are vast, it is useful to a beginner to get some notions and
concepts in order to progress easily. I tried to simplify the explanations as
much as I could to make them understandable by all, while giving useful
details, sometimes with tables.
R is a system for statistical analyses and graphics created by Ross Ihaka
and Robert Gentleman1 . R is both a software and a language considered as a
dialect of the S language created by the AT&T Bell Laboratories. S is available
as the software S-PLUS commercialized by Insightful 2 . There are important
differences in the designs of R and of S: those who want to know more on this
point can read the paper by Ihaka & Gentleman (1996) or the R-FAQ 3 , a copy
of which is also distributed with R.
R is freely distributed under the terms of the GNU General Public Licence 4 ;
its development and distribution are carried out by several statisticians known
as the R Development Core Team.
R is available in several forms: the sources (written mainly in C and
some routines in Fortran), essentially for Unix and Linux machines, or some
pre-compiled binaries for Windows, Linux, and Macintosh. The files needed
to install R, either from the sources or from the pre-compiled binaries, are
distributed from the internet site of the Comprehensive R Archive Network
(CRAN)5 where the instructions for the installation are also available. Regarding the distributions of Linux (Debian, . . . ), the binaries are generally
available for the most recent versions; look at the CRAN site if necessary.
R has many functions for statistical analyses and graphics; the latter are
visualized immediately in their own window and can be saved in various formats (jpg, png, bmp, ps, pdf, emf, pictex, xfig; the available formats may
depend on the operating system). The results from a statistical analysis are
displayed on the screen, some intermediate results (P-values, regression coefficients, residuals, . . . ) can be saved, written in a file, or used in subsequent
analyses.
The R language allows the user, for instance, to program loops to successively analyse several data sets. It is also possible to combine in a single
program different statistical functions to perform more complex analyses. The
1
Ihaka R. & Gentleman R. 1996. R: a language for data analysis and graphics. Journal
of Computational and Graphical Statistics 5: 299–314.
2
See http://www.insightful.com/products/splus/default.asp for more information
3
http://cran.r-project.org/doc/FAQ/R-FAQ.html
4
For more information: http://www.gnu.org/
5
http://cran.r-project.org/
1
R users may benefit from a large number of programs written for S and available on the internet6 , most of these programs can be used directly with R.
At first, R could seem too complex for a non-specialist. This may not
be true actually. In fact, a prominent feature of R is its flexibility. Whereas
a classical software displays immediately the results of an analysis, R stores
these results in an “object”, so that an analysis can be done with no result
displayed. The user may be surprised by this, but such a feature is very useful.
Indeed, the user can extract only the part of the results which is of interest.
For example, if one runs a series of 20 regressions and wants to compare the
different regression coefficients, R can display only the estimated coefficients:
thus the results may take a single line, whereas a classical software could well
open 20 results windows. We will see other examples illustrating the flexibility
of a system such as R compared to traditional softwares.
6
For example: http://stat.cmu.edu/S/
2
2
A few concepts before starting
Once R is installed on your computer, the software is executed by launching
the corresponding executable. The prompt, by default ‘>’, indicates that R
is waiting for your commands. Under Windows using the program Rgui.exe,
some commands (accessing the on-line help, opening files, . . . ) can be executed
via the pull-down menus. At this stage, a new user is likely to wonder “What
do I do now?” It is indeed very useful to have a few ideas on how R works
when it is used for the first time, and this is what we will see now.
We shall see first briefly how R works. Then, I will describe the “assign”
operator which allows creating objects, how to manage objects in memory,
and finally how to use the on-line help which is very useful when running R.
2.1
How R works
The fact that R is a language may deter some users who think “I can’t program”. This should not be the case for two reasons. First, R is an interpreted
language, not a compiled one, meaning that all commands typed on the keyboard are directly executed without requiring to build a complete program
like in most computer languages (C, Fortran, Pascal, . . . ).
Second, R’s syntax is very simple and intuitive. For instance, a linear
regression can be done with the command lm(y ~ x) which means “fitting
a linear model with y as response and x as predictor”. In R, in order to
be executed, a function always needs to be written with parentheses, even
if there is nothing within them (e.g., ls()). If one just types the name of a
function without parentheses, R will display the content of the function. In this
document, the names of the functions are generally written with parentheses in
order to distinguish them from other objects, unless the text indicates clearly
so.
When R is running, variables, data, functions, results, etc, are stored in
the active memory of the computer in the form of objects which have a name.
The user can do actions on these objects with operators (arithmetic, logical,
comparison, . . . ) and functions (which are themselves objects). The use of
operators is relatively intuitive, we will see the details later (p. 25). An R
function may be sketched as follows:
arguments −→
options −→
function
↑
default arguments
=⇒result
The arguments can be objects (“data”, formulae, expressions, . . . ), some
3
of which could be defined by default in the function; these default values may
be modified by the user by specifying options. An R function may require no
argument: either all arguments are defined by default (and their values can be
modified with the options), or no argument has been defined in the function.
We will see later in more details how to use and build functions (p. 67). The
present description is sufficient for the moment to understand how R works.
All the actions of R are done on objects stored in the active memory of
the computer: no temporary files are used (Fig. 1). The readings and writings
of files are used for input and output of data and results (graphics, . . . ). The
user executes the functions via some commands. The results are displayed
directly on the screen, stored in an object, or written on the disk (particularly
for graphics). Since the results are themselves objects, they can be considered
as data and analysed as such. Data files can be read from the local disk or
from a remote server through internet.
keyboard
mouse
commands
.../library/base/
/stast/
/graphics/
...
- functions and operators ?
“data” objects
)
screen
-
XXX
6
XXX
z
X
?
“results” objects
PS
Active memory
library of
functions
data
files
internet
JPEG
...
Hard disk
Figure 1: A schematic view of how R works.
The functions available to the user are stored in a library localised on
the disk in a directory called R HOME/library (R HOME is the directory
where R is installed). This directory contains packages of functions, which are
themselves structured in directories. The package named base is in a way the
core of R and contains the basic functions of the language, particularly, for
reading and manipulating data. Each package has a directory called R with
a file named like the package (for instance, for the package base, this is the
file R HOME/library/base/R/base). This file contains all the functions of the
package.
One of the simplest commands is to type the name of an object to display
its content. For instance, if an object n contents the value 10:
> n
[1] 10
4
The digit 1 within brackets indicates that the display starts at the first
element of n. This command is an implicit use of the function print and the
above example is similar to print(n) (in some situations, the function print
must be used explicitly, such as within a function or a loop).
The name of an object must start with a letter (A–Z and a–z) and can
include letters, digits (0–9), dots (.), and underscores ( ). R discriminates
between uppercase letters and lowercase ones in the names of the objects, so
that x and X can name two distinct objects (even under Windows).
2.2
Creating, listing and deleting the objects in memory
An object can be created with the “assign” operator which is written as an
arrow with a minus sign and a bracket; this symbol can be oriented left-to-right
or the reverse:
> n
> n
[1]
> 5
> n
[1]
> x
> X
> x
[1]
> X
[1]
<- 15
15
-> n
5
<- 1
<- 10
1
10
If the object already exists, its previous value is erased (the modification
affects only the objects in the active memory, not the data on the disk). The
value assigned this way may be the result of an operation and/or a function:
> n
> n
[1]
> n
> n
[1]
<- 10 + 2
12
<- 3 + rnorm(1)
2.208807
The function rnorm(1) generates a normal random variate with mean zero
and variance unity (p. 17). Note that you can simply type an expression
without assigning its value to an object, the result is thus displayed on the
screen but is not stored in memory:
> (10 + 2) * 5
[1] 60
5
The assignment will be omitted in the examples if not necessary for understanding.
The function ls lists simply the objects in memory: only the names of the
objects are displayed.
> name <- "Carmen"; n1 <- 10; n2 <- 100; m <- 0.5
> ls()
[1] "m"
"n1"
"n2"
"name"
Note the use of the semi-colon to separate distinct commands on the same
line. If we want to list only the objects which contain a given character in
their name, the option pattern (which can be abbreviated with pat) can be
used:
> ls(pat = "m")
[1] "m"
"name"
To restrict the list of objects whose names start with this character:
> ls(pat = "^m")
[1] "m"
The function ls.str displays some details on the objects in memory:
> ls.str()
m : num 0.5
n1 : num 10
n2 : num 100
name : chr "Carmen"
The option pattern can be used in the same way as with ls. Another
useful option of ls.str is max.level which specifies the level of detail for the
display of composite objects. By default, ls.str displays the details of all
objects in memory, included the columns of data frames, matrices and lists,
which can result in a very long display. We can avoid to display all these
details with the option max.level = -1:
> M <- data.frame(n1, n2, m)
> ls.str(pat = "M")
M : ‘data.frame’:
1 obs. of
$ n1: num 10
$ n2: num 100
$ m : num 0.5
> ls.str(pat="M", max.level=-1)
M : ‘data.frame’:
1 obs. of
3 variables:
3 variables:
To delete objects in memory, we use the function rm: rm(x) deletes the
object x, rm(x,y) deletes both the objects x et y, rm(list=ls()) deletes all
the objects in memory; the same options mentioned for the function ls() can
then be used to delete selectively some objects: rm(list=ls(pat="^m")).
6
2.3
The on-line help
The on-line help of R gives very useful information on how to use the functions.
Help is available directly for a given function, for instance:
> ?lm
will display, within R, the help page for the function lm() (linear model). The
commands help(lm) and help("lm") have the same effect. The last one must
be used to access help with non-conventional characters:
> ?*
Error: syntax error
> help("*")
Arithmetic
package:base
R Documentation
Arithmetic Operators
...
Calling help opens a page (this depends on the operating system) with
general information on the first line such as the name of the package where
is (are) the documented function(s) or operators. Then comes a title followed
by sections which give detailed information.
Description: brief description.
Usage: for a function, gives the name with all its arguments and the possible
options (with the corresponding default values); for an operator gives
the typical use.
Arguments: for a function, details each of its arguments.
Details: detailed description.
Value: if applicable, the type of object returned by the function or the operator.
See Also: other help pages close or similar to the present one.
Examples: some examples which can generally be executed without opening
the help with the function example.
For beginners, it is good to look at the section Examples. Generally, it
is useful to read carefully the section Arguments. Other sections may be
encountered, such as Note, References or Author(s).
By default, the function help only searches in the packages which are
loaded in memory. The option try.all.packages, which default is FALSE,
allows to search in all packages if its value is TRUE:
7
> help("bs")
No documentation for ’bs’ in specified packages and libraries:
you could try ’help.search("bs")’
> help("bs", try.all.packages = TRUE)
Help for topic ’bs’ is not in any loaded package but
can be found in the following packages:
Package
splines
Library
/usr/lib/R/library
Note that in this case the help page of the function bs is not displayed.
The user can display help pages from a package not loaded in memory using
the option package:
> help("bs", package = "splines")
bs
package:splines
R Documentation
B-Spline Basis for Polynomial Splines
Description:
Generate the B-spline basis matrix for a polynomial spline.
...
The help in html format (read, e.g., with Netscape) is called by typing:
> help.start()
A search with keywords is possible with this html help. The section See
Also has here hypertext links to other function help pages. The search with
keywords is also possible in R with the function help.search. The latter
looks for a specified topic, given as a character string, in the help pages of all
installed packages. For instance, help.search("tree") will display a list of
the functions which help pages mention “tree”. Note that if some packages
have been recently installed, it may be useful to refresh the database used by
help.search using the option rebuild (e.g., help.search("tree", rebuild
= TRUE)).
The fonction apropos finds all functions which name contains the character
string given as argument; only the packages loaded in memory are searched:
> apropos(help)
[1] "help"
[4] "help.start"
".helpForCall" "help.search"
8
3
3.1
Data with R
Objects
We have seen that R works with objects which are, of course, characterized by
their names and their content, but also by attributes which specify the kind of
data represented by an object. In order to understand the usefulness of these
attributes, consider a variable that takes the value 1, 2, or 3: such a variable
could be an integer variable (for instance, the number of eggs in a nest), or
the coding of a categorical variable (for instance, sex in some populations of
crustaceans: male, female, or hermaphrodite).
It is clear that the statistical analysis of this variable will not be the same in
both cases: with R, the attributes of the object give the necessary information.
More technically, and more generally, the action of a function on an object
depends on the attributes of the latter.
All objects have two intrinsic attributes: mode and length. The mode
is the basic type of the elements of the object; there are four main modes:
numeric, character, complex7 , and logical (FALSE or TRUE). Other modes exist
but they do not represent data, for instance function or expression. The length
is the number of elements of the object. To display the mode and the length
of an object, one can use the functions mode and length, respectively:
> x <- 1
> mode(x)
[1] "numeric"
> length(x)
[1] 1
> A <- "Gomphotherium"; compar <- TRUE; z <- 1i
> mode(A); mode(compar); mode(z)
[1] "character"
[1] "logical"
[1] "complex"
Whatever the mode, missing data are represented by NA (not available).
A very large numeric value can be specified with an exponential notation:
> N <- 2.1e23
> N
[1] 2.1e+23
R correctly represents non-finite numeric values, such as ±∞ with Inf and
-Inf, or values which are not numbers with NaN (not a number ).
7
The mode complex will not be discussed in this document.
9
> x <- 5/0
> x
[1] Inf
> exp(x)
[1] Inf
> exp(-x)
[1] 0
> x - x
[1] NaN
A value of mode character is input with double quotes ". It is possible
to include this latter character in the value if it follows a backslash \. The
two charaters altogether \" will be treated in a specific way by some functions
such as cat for display on screen, or write.table to write on the disk (p. 14,
the option qmethod of this function).
> x <- "Double quotes \" delimitate R’s strings."
> x
[1] "Double quotes \" delimitate R’s strings."
> cat(x)
Double quotes " delimitate R’s strings.
Alternatively, variables of mode character can be delimited with single
quotes (’); in this case it is not necessary to escape double quotes with backslashes (but single quotes must be!):
> x <- ’Double quotes " delimitate R\’s strings.’
> x
[1] "Double quotes \" delimitate R’s strings."
The following table gives an overview of the type of objects representing
data.
object
vector
factor
array
matrix
data frame
ts
list
modes
several modes
possible in the
same object?
numeric, character, complex or logical
numeric or character
numeric, character, complex or logical
numeric, character, complex or logical
numeric, character, complex or logical
numeric, character, complex or logical
numeric, character, complex, logical,
function, expression, . . .
No
No
No
No
Yes
No
Yes
10
A vector is a variable in the commonly admitted meaning. A factor is a
categorical variable. An array is a table with k dimensions, a matrix being
a particular case of array with k = 2. Note that the elements of an array
or of a matrix are all of the same mode. A data frame is a table composed
with one or several vectors and/or factors all of the same length but possibly
of different modes. A ‘ts’ is a time series data set and so contains additional
attributes such as frequency and dates. Finally, a list can contain any type of
object, included lists!
For a vector, its mode and length are sufficient to describe the data. For
other objects, other information is necessary and it is given by non-intrinsic
attributes. Among these attributes, we can cite dim which corresponds to the
dimensions of an object. For example, a matrix with 2 lines and 2 columns
has for dim the pair of values [2, 2], but its length is 4.
3.2
Reading data in a file
For reading and writing in files, R uses the working directory. To find this
directory, the command getwd() (get working directory) can be used, and the
working directory can be changed with setwd("C:/data") or setwd("/home/paradis/R"). It is necessary to give the path to a file if it is not in the working
directory.8
R can read data stored in text (ASCII) files with the following functions:
read.table (which has several variants, see below), scan and read.fwf. R
can also read files in other formats (Excel, SAS, SPSS, . . . ), and access SQLtype databases, but the functions needed for this are not in the package base.
These functionalities are very useful for a more advanced use of R, but we will
restrict here to reading files in ASCII format.
The function read.table has for effect to create a data frame, and so is
the main way to read data in tabular form. For instance, if one has a file
named data.dat, the command:
> mydata <- read.table("data.dat")
will create a data frame named mydata, and each variable will be named, by default, V1, V2, . . . and can be accessed individually by mydata$V1, mydata$V2,
. . . , or by mydata["V1"], mydata["V2"], . . . , or, still another solution, by
mydata[, 1], mydata[,2 ], . . . 9 There are several options whose default
values (i.e. those used by R if they are omitted by the user) are detailed in
the following table:
read.table(file, header = FALSE, sep = "", quote = "\"’", dec = ".",
8
Under Windows, it is useful to create a short-cut of Rgui.exe then edit its properties
and change the directory in the field “Start in:” under the tab “Short-cut”: this directory
will then be the working directory if R is started from this short-cut.
9
There is a difference: mydata$V1 and mydata[, 1] are vectors whereas mydata["V1"] is
a data frame. We will see later (p. 18) some details on manipulating objects.
11
row.names, col.names, as.is = FALSE, na.strings = "NA",
colClasses = NA, nrows = -1,
skip = 0, check.names = TRUE, fill = !blank.lines.skip,
strip.white = FALSE, blank.lines.skip = TRUE,
comment.char = "#")
file
header
sep
quote
dec
row.names
col.names
as.is
na.strings
colClasses
nrows
skip
check.names
fill
strip.white
blank.lines.skip
comment.char
the name of the file (within "" or a variable of mode character),
possibly with its path (the symbol \ is not allowed and must be
replaced by /, even under Windows), or a remote access to a file of
type URL (http://...)
a logical (FALSE or TRUE) indicating if the file contains the names of
the variables on its first line
the field separator used in the file, for instance sep="\t" if it is a
tabulation
the characters used to cite the variables of mode character
the character used for the decimal point
a vector with the names of the lines which can be either a vector of
mode character, or the number (or the name) of a variable of the
file (by default: 1, 2, 3, . . . )
a vector with the names of the variables (by default: V1, V2, V3,
...)
controls the conversion of character variables as factors (if FALSE)
or keeps them as characters (TRUE); as.is can be a logical, numeric
or character vector specifying the variables to be kept as character
the value given to missing data (converted as NA)
a vector of mode character giving the classes to attribute to the
columns
the maximum number of lines to read (negative values are ignored)
the number of lines to be skipped before reading the data
if TRUE, checks that the variable names are valid for R
if TRUE and all lines do not have the same number of variables,
“blanks” are added
(conditional to sep) if TRUE, deletes extra spaces before and after
the character variables
if TRUE, ignores “blank” lines
a character defining comments in the data file, the rest of the
line after this character is ignored (to disable this argument, use
comment.char = "")
The variants of read.table are useful since they have different default
values:
read.csv(file, header = TRUE, sep = ",", quote="\"", dec=".",
fill = TRUE, ...)
read.csv2(file, header = TRUE, sep = ";", quote="\"", dec=",",
fill = TRUE, ...)
read.delim(file, header = TRUE, sep = "\t", quote="\"", dec=".",
fill = TRUE, ...)
read.delim2(file, header = TRUE, sep = "\t", quote="\"", dec=",",
fill = TRUE, ...)
12
The function scan is more flexible than read.table. A difference is that
it is possible to specify the mode of the variables, for example:
> mydata <- scan("data.dat", what = list("", 0, 0))
reads in the file data.dat three variables, the first is of mode character and the
next two are of mode numeric. Another important distinction is that scan()
can be used to create different objects, vectors, matrices, data frames, lists,
. . . In the above example, mydata is a list of three vectors. By default, that is
if what is omitted, scan() creates a numeric vector. If the data read do not
correspond to the mode(s) expected (either by default, or specified by what),
an error message is returned. The options are the followings.
scan(file = "", what = double(0), nmax = -1, n = -1, sep = "",
quote = if (sep=="\n") "" else "’\"", dec = ".",
skip = 0, nlines = 0, na.strings = "NA",
flush = FALSE, fill = FALSE, strip.white = FALSE, quiet = FALSE,
blank.lines.skip = TRUE, multi.line = TRUE, comment.char = "",
allowEscapes = TRUE)
file
what
nmax
n
sep
quote
dec
skip
nlines
na.string
flush
fill
strip.white
quiet
blank.lines.skip
multi.line
comment.char
allowEscapes
the name of the file (within ""), possibly with its path (the symbol
\ is not allowed and must be replaced by /, even under Windows),
or a remote access to a file of type URL (http://...); if file="", the
data are entered with the keyboard (the entree is terminated by a
blank line)
specifies the mode(s) of the data (numeric by default)
the number of data to read, or, if what is a list, the number of lines
to read (by default, scan reads the data up to the end of file)
the number of data to read (by default, no limit)
the field separator used in the file
the characters used to cite the variables of mode character
the character used for the decimal point
the number of lines to be skipped before reading the data
the number of lines to read
the value given to missing data (converted as NA)
a logical, if TRUE, scan goes to the next line once the number of
columns has been reached (allows the user to add comments in the
data file)
if TRUE and all lines do not have the same number of variables,
“blanks” are added
(conditional to sep) if TRUE, deletes extra spaces before and after
the character variables
a logical, if FALSE, scan displays a line showing which fields have
been read
if TRUE, ignores blank lines
if what is a list, specifies if the variables of the same individual are
on a single line in the file (FALSE)
a character defining comments in the data file, the rest of the line
after this character is ignored (the default is to have this disabled)
specifies whether C-style escapes (e.g., ‘\t’) be processed (the default) or read as verbatim
13
The function read.fwf can be used to read in a file some data in fixed
width format:
read.fwf(file, widths, header = FALSE, sep = "\t",
as.is = FALSE, skip = 0, row.names, col.names,
n = -1, buffersize = 2000, ...)
The options are the same than for read.table() except widths which specifies the width of the fields
(buffersize is the maximum number of lines read simultaneously). For example, if a file named data.txt has
the data indicated on the right, one can read the data
with the following command:
A1.501.2
A1.551.3
B1.601.4
B1.651.5
C1.701.6
C1.751.7
> mydata <- read.fwf("data.txt", widths=c(1, 4, 3))
> mydata
V1
V2 V3
1 A 1.50 1.2
2 A 1.55 1.3
3 B 1.60 1.4
4 B 1.65 1.5
5 C 1.70 1.6
6 C 1.75 1.7
3.3
Saving data
The function write.table writes in a file an object, typically a data frame but
this could well be another kind of object (vector, matrix, . . . ). The arguments
and options are:
write.table(x, file = "", append = FALSE, quote = TRUE, sep = " ",
eol = "\n", na = "NA", dec = ".", row.names = TRUE,
col.names = TRUE, qmethod = c("escape", "double"))
x
file
append
quote
sep
eol
na
dec
row.names
col.names
qmethod
the name of the object to be written
the name of the file (by default the object is displayed on the screen)
if TRUE adds the data without erasing those possibly existing in the file
a logical or a numeric vector: if TRUE the variables of mode character and
the factors are written within "", otherwise the numeric vector indicates
the numbers of the variables to write within "" (in both cases the names
of the variables are written within "" but not if quote = FALSE)
the field separator used in the file
the character to be used at the end of each line ("\n" is a carriage-return)
the character to be used for missing data
the character used for the decimal point
a logical indicating whether the names of the lines are written in the file
id. for the names of the columns
specifies, if quote=TRUE, how double quotes " included in variables of mode
character are treated: if "escape" (or "e", the default) each " is replaced
by \", if "d" each " is replaced by ""
14
To write in a simpler way an object in a file, the command write(x,
file="data.txt") can be used, where x is the name of the object (which
can be a vector, a matrix, or an array). There are two options: nc (or ncol)
which defines the number of columns in the file (by default nc=1 if x is of mode
character, nc=5 for the other modes), and append (a logical) to add the data
without deleting those possibly already in the file (TRUE) or deleting them if
the file already exists (FALSE, the default).
To record a group of objects of any type, we can use the command save(x,
y, z, file= "xyz.RData"). To ease the transfert of data between different machines, the option ascii = TRUE can be used. The data (which are
now called a workspace in R’s jargon) can be loaded later in memory with
load("xyz.RData"). The function save.image() is a short-cut for save(list
=ls(all=TRUE), file=".RData").
3.4
Generating data
3.4.1
Regular sequences
A regular sequence of integers, for example from 1 to 30, can be generated
with:
> x <- 1:30
The resulting vector x has 30 elements. The operator ‘:’ has priority on the
arithmetic operators within an expression:
> 1:10-1
[1] 0 1 2 3 4 5 6 7 8 9
> 1:(10-1)
[1] 1 2 3 4 5 6 7 8 9
The function seq can generate sequences of real numbers as follows:
> seq(1, 5, 0.5)
[1] 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
where the first number indicates the beginning of the sequence, the second one
the end, and the third one the increment to be used to generate the sequence.
One can use also:
> seq(length=9, from=1, to=5)
[1] 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
One can also type directly the values using the function c:
> c(1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5)
[1] 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
15
It is also possible, if one wants to enter some data on the keyboard, to use
the function scan with simply the default options:
> z <- scan()
1: 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
10:
Read 9 items
> z
[1] 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
The function rep creates a vector with all its elements identical:
> rep(1, 30)
[1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
The function sequence creates a series of sequences of integers each ending
by the numbers given as arguments:
> sequence(4:5)
[1] 1 2 3 4 1 2 3 4 5
> sequence(c(10,5))
[1] 1 2 3 4 5 6
7
8
9 10
1
2
3
4
5
The function gl (generate levels) is very useful because it generates regular
series of factors. The usage of this fonction is gl(k, n) where k is the number
of levels (or classes), and n is the number of replications in each level. Two
options may be used: length to specify the number of data produced, and
labels to specify the names of the levels of the factor. Examples:
> gl(3, 5)
[1] 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3
Levels: 1 2 3
> gl(3, 5, length=30)
[1] 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3
Levels: 1 2 3
> gl(2, 6, label=c("Male", "Female"))
[1] Male
Male
Male
Male
Male
Male
[7] Female Female Female Female Female Female
Levels: Male Female
> gl(2, 10)
[1] 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2
Levels: 1 2
> gl(2, 1, length=20)
[1] 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2
Levels: 1 2
> gl(2, 2, length=20)
[1] 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2
Levels: 1 2
16
Finally, expand.grid() creates a data frame with all combinations of vectors or factors given as arguments:
> expand.grid(h=c(60,80), w=c(100, 300), sex=c("Male", "Female"))
h
w
sex
1 60 100
Male
2 80 100
Male
3 60 300
Male
4 80 300
Male
5 60 100 Female
6 80 100 Female
7 60 300 Female
8 80 300 Female
3.4.2
Random sequences
law
function
Gaussian (normal)
exponential
gamma
Poisson
Weibull
Cauchy
beta
‘Student’ (t)
Fisher–Snedecor (F )
Pearson (χ2 )
binomial
multinomial
geometric
hypergeometric
logistic
lognormal
negative binomial
uniform
Wilcoxon’s statistics
rnorm(n, mean=0, sd=1)
rexp(n, rate=1)
rgamma(n, shape, scale=1)
rpois(n, lambda)
rweibull(n, shape, scale=1)
rcauchy(n, location=0, scale=1)
rbeta(n, shape1, shape2)
rt(n, df)
rf(n, df1, df2)
rchisq(n, df)
rbinom(n, size, prob)
rmultinom(n, size, prob)
rgeom(n, prob)
rhyper(nn, m, n, k)
rlogis(n, location=0, scale=1)
rlnorm(n, meanlog=0, sdlog=1)
rnbinom(n, size, prob)
runif(n, min=0, max=1)
rwilcox(nn, m, n), rsignrank(nn, n)
It is useful in statistics to be able to generate random data, and R can
do it for a large number of probability density functions. These functions are
of the form rfunc (n, p1, p2, ...), where func indicates the probability
distribution, n the number of data generated, and p1, p2, . . . are the values of
the parameters of the distribution. The above table gives the details for each
distribution, and the possible default values (if none default value is indicated,
this means that the parameter must be specified by the user).
Most of these functions have counterparts obtained by replacing the letter
r with d, p or q to get, respectively, the probability density (dfunc (x, ...)),
17
the cumulative probability density (pfunc (x, ...)), and the value of quantile
(qfunc (p, ...), with 0 < p < 1). The last two series of functions can be
used to find critical values or P -values of statistical tests. For instance, the
critical values for a two-tailed test following a normal distribution at the 5%
threshold are:
> qnorm(0.025)
[1] -1.959964
> qnorm(0.975)
[1] 1.959964
For the one-tailed version of the same test, either qnorm(0.05) or 1 qnorm(0.95) will be used depending on the form of the alternative hypothesis.
The P -value of a test, say χ2 = 3.84 with df = 1, is:
> 1 - pchisq(3.84, 1)
[1] 0.05004352
3.5
3.5.1
Manipulating objects
Creating objects
We have seen previously different ways to create objects using the assign operator; the mode and the type of objects so created are generally determined
implicitly. It is possible to create an object and specifying its mode, length,
type, etc. This approach is interesting in the perspective of manipulating objects. One can, for instance, create an ‘empty’ object and then modify its
elements successively which is more efficient than putting all its elements together with c(). The indexing system could be used here, as we will see later
(p. 26).
It can also be very convenient to create objects from others. For example,
if one wants to fit a series of models, it is simple to put the formulae in a list,
and then to extract the elements successively to insert them in the function
lm.
At this stage of our learning of R, the interest in learning the following
functionalities is not only practical but also didactic. The explicit construction
of objects gives a better understanding of their structure, and allows us to go
further in some notions previously mentioned.
Vector. The function vector, which has two arguments mode and length,
creates a vector which elements have a value depending on the mode
specified as argument: 0 if numeric, FALSE if logical, or "" if character. The following functions have exactly the same effect and have for
single argument the length of the vector: numeric(), logical(), and
character().
18
Factor. A factor includes not only the values of the corresponding categorical
variable, but also the different possible levels of that variable (even if they
are not present in the data). The function factor creates a factor with
the following options:
factor(x, levels = sort(unique(x), na.last = TRUE),
labels = levels, exclude = NA, ordered = is.ordered(x))
levels specifies the possible levels of the factor (by default the unique
values of the vector x), labels defines the names of the levels, exclude
the values of x to exclude from the levels, and ordered is a logical
argument specifying whether the levels of the factor are ordered. Recall
that x is of mode numeric or character. Some examples follow.
> factor(1:3)
[1] 1 2 3
Levels: 1 2 3
> factor(1:3, levels=1:5)
[1] 1 2 3
Levels: 1 2 3 4 5
> factor(1:3, labels=c("A", "B", "C"))
[1] A B C
Levels: A B C
> factor(1:5, exclude=4)
[1] 1 2 3 NA 5
Levels: 1 2 3 5
The function levels extracts the possible levels of a factor:
> ff <- factor(c(2, 4), levels=2:5)
> ff
[1] 2 4
Levels: 2 3 4 5
> levels(ff)
[1] "2" "3" "4" "5"
Matrix. A matrix is actually a vector with an additional attribute (dim)
which is itself a numeric vector with length 2, and defines the numbers
of rows and columns of the matrix. A matrix can be created with the
function matrix:
matrix(data = NA, nrow = 1, ncol = 1, byrow = FALSE,
dimnames = NULL)
19
The option byrow indicates whether the values given by data must fill
successively the columns (the default) or the rows (if TRUE). The option
dimnames allows to give names to the rows and columns.
> matrix(data=5, nr=2, nc=2)
[,1] [,2]
[1,]
5
5
[2,]
5
5
> matrix(1:6, 2, 3)
[,1] [,2] [,3]
[1,]
1
3
5
[2,]
2
4
6
> matrix(1:6, 2, 3, byrow=TRUE)
[,1] [,2] [,3]
[1,]
1
2
3
[2,]
4
5
6
Another way to create a matrix is to give the appropriate values to the
dim attribute (which is initially NULL):
> x <- 1:15
> x
[1] 1 2 3 4 5
> dim(x)
NULL
> dim(x) <- c(5, 3)
> x
[,1] [,2] [,3]
[1,]
1
6
11
[2,]
2
7
12
[3,]
3
8
13
[4,]
4
9
14
[5,]
5
10
15
6
7
8
9 10 11 12 13 14 15
Data frame. We have seen that a data frame is created implicitly by the
function read.table; it is also possible to create a data frame with the
function data.frame. The vectors so included in the data frame must
be of the same length, or if one of the them is shorter, it is “recycled” a
whole number of times:
> x <- 1:4; n <- 10; M <- c(10, 35); y <- 2:4
> data.frame(x, n)
x n
1 1 10
2 2 10
20
3 3 10
4 4 10
> data.frame(x, M)
x M
1 1 10
2 2 35
3 3 10
4 4 35
> data.frame(x, y)
Error in data.frame(x, y) :
arguments imply differing number of rows: 4, 3
If a factor is included in a data frame, it must be of the same length than
the vector(s). It is possible to change the names of the columns with,
for instance, data.frame(A1=x, A2=n). One can also give names to the
rows with the option row.names which must be, of course, a vector of
mode character and of length equal to the number of lines of the data
frame. Finally, note that data frames have an attribute dim similarly to
matrices.
List. A list is created in a way similar to data frames with the function list.
There is no constraint on the objects that can be included. In contrast
to data.frame(), the names of the objects are not taken by default;
taking the vectors x and y of the previous example:
> L1 <- list(x, y); L2 <- list(A=x, B=y)
> L1
[[1]]
[1] 1 2 3 4
[[2]]
[1] 2 3 4
> L2
$A
[1] 1 2 3 4
$B
[1] 2 3 4
> names(L1)
NULL
> names(L2)
[1] "A" "B"
Time-series. The function ts creates an object of class "ts" from a vector
(single time-series) or a matrix (multivariate time-series), and some op21
tions which characterize the series. The options, with the default values,
are:
ts(data = NA, start = 1, end = numeric(0), frequency = 1,
deltat = 1, ts.eps = getOption("ts.eps"), class, names)
data
start
end
frequency
deltat
ts.eps
class
names
a vector or a matrix
the time of the first observation, either a number, or a
vector of two integers (see the examples below)
the time of the last observation specified in the same way
than start
the number of observations per time unit
the fraction of the sampling period between successive
observations (ex. 1/12 for monthly data); only one of
frequency or deltat must be given
tolerance for the comparison of series. The frequencies
are considered equal if their difference is less than ts.eps
class to give to the object; the default is "ts" for a single
series, and c("mts", "ts") for a multivariate series
a vector of mode character with the names of the individual series in the case of a multivariate series; by default
the names of the columns of data, or Series 1, Series
2, . . .
A few examples of time-series created with ts:
> ts(1:10, start = 1959)
Time Series:
Start = 1959
End = 1968
Frequency = 1
[1] 1 2 3 4 5 6 7 8 9 10
> ts(1:47, frequency = 12, start = c(1959, 2))
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1959
1
2
3
4
5
6
7
8
9 10 11
1960 12 13 14 15 16 17 18 19 20 21 22 23
1961 24 25 26 27 28 29 30 31 32 33 34 35
1962 36 37 38 39 40 41 42 43 44 45 46 47
> ts(1:10, frequency = 4, start = c(1959, 2))
Qtr1 Qtr2 Qtr3 Qtr4
1959
1
2
3
1960
4
5
6
7
1961
8
9
10
> ts(matrix(rpois(36, 5), 12, 3), start=c(1961, 1), frequency=12)
Series 1 Series 2 Series 3
22
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
1961
1961
1961
1961
1961
1961
1961
1961
1961
1961
1961
1961
8
6
2
8
4
4
4
11
6
6
5
8
5
6
3
5
9
6
2
6
5
5
5
5
4
9
3
4
3
13
6
4
7
7
7
2
Expression. The objects of mode expression have a fundamental role in R.
An expression is a series of characters which makes sense for R. All valid
commands are expressions. When a command is typed directly on the
keyboard, it is then evaluated by R and executed if it is valid. In many
circumstances, it is useful to construct an expression without evaluating
it: this is what the function expression is made for. It is, of course,
possible to evaluate the expression subsequently with eval().
> x <- 3; y <- 2.5; z <- 1
> exp1 <- expression(x / (y + exp(z)))
> exp1
expression(x/(y + exp(z)))
> eval(exp1)
[1] 0.5749019
Expressions can be used, among other things, to include equations in
graphs (p. 42). An expression can be created from a variable of mode
character. Some functions take expressions as arguments, for example D
which returns partial derivatives:
> D(exp1, "x")
1/(y + exp(z))
> D(exp1, "y")
-x/(y + exp(z))^2
> D(exp1, "z")
-x * exp(z)/(y + exp(z))^2
3.5.2
Converting objects
The reader has surely realized that the differences between some types of
objects are small; it is thus logical that it is possible to convert an object from
a type to another by changing some of its attributes. Such a conversion will be
done with a function of the type as.something . R (version 2.1.0) has, in the
23
packages base and utils, 98 of such functions, so we will not go in the deepest
details here.
The result of a conversion depends obviously of the attributes of the converted object. Genrally, conversion follows intuitive rules. For the conversion
of modes, the following table summarizes the situation.
Conversion to
Function
numeric
as.numeric
logical
as.logical
character
as.character
Rules
FALSE →
TRUE →
"1", "2", . . . →
"A", . . . →
0→
other numbers →
"FALSE", "F" →
"TRUE", "T" →
other characters →
1, 2, . . . →
FALSE →
TRUE →
0
1
1, 2, . . .
NA
FALSE
TRUE
FALSE
TRUE
NA
"1", "2", . . .
"FALSE"
"TRUE"
There are functions to convert the types of objects (as.matrix, as.ts,
as.data.frame, as.expression, . . . ). These functions will affect attributes
other than the modes during the conversion. The results are, again, generally
intuitive. A situation frequently encountered is the conversion of factors into
numeric values. In this case, R does the conversion with the numeric coding
of the levels of the factor:
> fac <- factor(c(1, 10))
> fac
[1] 1 10
Levels: 1 10
> as.numeric(fac)
[1] 1 2
This makes sense when considering a factor of mode character:
> fac2 <- factor(c("Male", "Female"))
> fac2
[1] Male
Female
Levels: Female Male
> as.numeric(fac2)
[1] 2 1
Note that the result is not NA as may have been expected from the table
above.
24
To convert a factor of mode numeric into a numeric vector but keeping the
levels as they are originally specified, one must first convert into character,
then into numeric.
> as.numeric(as.character(fac))
[1] 1 10
This procedure is very useful if in a file a numeric variable has also nonnumeric values. We have seen that read.table() in such a situation will, by
default, read this column as a factor.
3.5.3
Operators
We have seen previously that there are three main types of operators in R 10 .
Here is the list.
Operators
Comparison
Arithmetic
+
*
/
^
%%
%/%
addition
subtraction
multiplication
division
power
modulo
integer division
<
>
<=
>=
==
!=
lesser than
greater than
lesser than or equal to
greater than or equal to
equal
different
Logical
! x
x & y
x && y
x | y
x || y
xor(x, y)
logical NOT
logical AND
id.
logical OR
id.
exclusive OR
The arithmetic and comparison operators act on two elements (x + y, a
< b). The arithmetic operators act not only on variables of mode numeric or
complex, but also on logical variables; in this latter case, the logical values
are coerced into numeric. The comparison operators may be applied to any
mode: they return one or several logical values.
The logical operators are applied to one (!) or two objects of mode logical,
and return one (or several) logical values. The operators “AND” and “OR”
exist in two forms: the single one operates on each elements of the objects and
returns as many logical values as comparisons done; the double one operates
on the first element of the objects.
It is necessary to use the operator “AND” to specify an inequality of the
type 0 < x < 1 which will be coded with: 0 < x & x < 1. The expression 0
< x < 1 is valid, but will not return the expected result: since both operators
are the same, they are executed successively from left to right. The comparison
0 < x is first done and returns a logical value which is then compared to 1
(TRUE or FALSE < 1): in this situation, the logical value is implicitly coerced
into numeric (1 or 0 < 1).
10
The following characters are also operators for R: $, @, [, [[, :, ?, <-, <<-, =, ::. A
table of operators describing precedence rules can be found with ?Syntax.
25
> x <- 0.5
> 0 < x < 1
[1] FALSE
The comparison operators operate on each element of the two objects
being compared (recycling the values of the shortest one if necessary), and
thus returns an object of the same size. To compare ‘wholly’ two objects, two
functions are available: identical and all.equal.
> x <- 1:3; y <- 1:3
> x == y
[1] TRUE TRUE TRUE
> identical(x, y)
[1] TRUE
> all.equal(x, y)
[1] TRUE
identical compares the internal representation of the data and returns
TRUE if the objects are strictly identical, and FALSE otherwise. all.equal
compares the “near equality” of two objects, and returns TRUE or display a
summary of the differences. The latter function takes the approximation of
the computing process into account when comparing numeric values. The
comparison of numeric values on a computer is sometimes surprising!
> 0.9 == (1 - 0.1)
[1] TRUE
> identical(0.9, 1 - 0.1)
[1] TRUE
> all.equal(0.9, 1 - 0.1)
[1] TRUE
> 0.9 == (1.1 - 0.2)
[1] FALSE
> identical(0.9, 1.1 - 0.2)
[1] FALSE
> all.equal(0.9, 1.1 - 0.2)
[1] TRUE
> all.equal(0.9, 1.1 - 0.2, tolerance = 1e-16)
[1] "Mean relative difference: 1.233581e-16"
3.5.4
Accessing the values of an object: the indexing system
The indexing system is an efficient and flexible way to access selectively the
elements of an object; it can be either numeric or logical. To access, for
example, the third value of a vector x, we just type x[3] which can be used
either to extract or to change this value:
> x <- 1:5
26
> x[3]
[1] 3
> x[3] <- 20
> x
[1] 1 2 20
4
5
The index itself can be a vector of mode numeric:
> i <- c(1, 3)
> x[i]
[1] 1 20
If x is a matrix or a data frame, the value of the ith line and j th column
is accessed with x[i, j]. To access all values of a given row or column, one
has simply to omit the appropriate index (without forgetting the comma!):
> x <- matrix(1:6, 2, 3)
> x
[,1] [,2] [,3]
[1,]
1
3
5
[2,]
2
4
6
> x[, 3] <- 21:22
> x
[,1] [,2] [,3]
[1,]
1
3
21
[2,]
2
4
22
> x[, 3]
[1] 21 22
You have certainly noticed that the last result is a vector and not a matrix.
The default behaviour of R is to return an object of the lowest dimension
possible. This can be altered with the option drop which default is TRUE:
> x[, 3, drop = FALSE]
[,1]
[1,]
21
[2,]
22
This indexing system is easily generalized to arrays, with as many indices
as the number of dimensions of the array (for example, a three dimensional
array: x[i, j, k], x[, , 3], x[, , 3, drop = FALSE], and so on). It may
be useful to keep in mind that indexing is made with square brackets, while
parentheses are used for the arguments of a function:
> x(1)
Error: couldn’t find function "x"
27
Indexing can also be used to suppress one or several rows or columns
using negative values. For example, x[-1, ] will suppress the first row, while
x[-c(1, 15), ] will do the same for the 1st and 15th rows. Using the matrix
defined above:
> x[, -1]
[,1] [,2]
[1,]
3
21
[2,]
4
22
> x[, -(1:2)]
[1] 21 22
> x[, -(1:2), drop = FALSE]
[,1]
[1,]
21
[2,]
22
For vectors, matrices and arrays, it is possible to access the values of an
element with a comparison expression as the index:
> x <- 1:10
> x[x >= 5]
> x
[1] 1 2
> x[x == 1]
> x
[1] 25 2
<- 20
3 4 20 20 20 20 20 20
<- 25
3
4 20 20 20 20 20 20
A practical use of the logical indexing is, for instance, the possibility to
select the even elements of an integer variable:
> x <- rpois(40, lambda=5)
> x
[1] 5 9 4 7 7 6 4 5 11 3
[21] 4 6 6 5 4 5 3 4 3 3
> x[x %% 2 == 0]
[1] 4 6 4 2 2 2 4 6 6 4 4 8 4 2 4
5
3
7
7
1
7
5
3
3
8
9
1
2
4
2
2
5
1
2
4
Thus, this indexing system uses the logical values returned, in the above
examples, by comparison operators. These logical values can be computed
beforehand, they then will be recycled if necessary:
> x <- 1:40
> s <- c(FALSE, TRUE)
> x[s]
[1] 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
28
Logical indexing can also be used with data frames, but with caution since
different columns of the data drame may be of different modes.
For lists, accessing the different elements (which can be any kind of object)
is done either with single or with double square brackets: the difference is
that with single brackets a list is returned, whereas double brackets extract the
object from the list. The result can then be itself indexed as previously seen for
vectors, matrices, etc. For instance, if the third object of a list is a vector, its
ith value can be accessed using my.list[[3]][i], if it is a three dimensional
array using my.list[[3]][i, j, k], and so on. Another difference is that
my.list[1:2] will return a list with the first and second elements of the
original list, whereas my.list[[1:2]] will no not give the expected result.
3.5.5
Accessing the values of an object with names
The names are labels of the elements of an object, and thus of mode character. They are generally optional attributes. There are several kinds of names
(names, colnames, rownames, dimnames).
The names of a vector are stored in a vector of the same length of the
object, and can be accessed with the function names.
> x <- 1:3
> names(x)
NULL
> names(x) <- c("a", "b", "c")
> x
a b c
1 2 3
> names(x)
[1] "a" "b" "c"
> names(x) <- NULL
> x
[1] 1 2 3
For matrices and data frames, colnames and rownames are labels of the
columns and rows, respectively. They can be accessed either with their respective functions, or with dimnames which returns a list with both vectors.
>
>
>
>
X <- matrix(1:4, 2)
rownames(X) <- c("a", "b")
colnames(X) <- c("c", "d")
X
c d
a 1 3
b 2 4
> dimnames(X)
[[1]]
[1] "a" "b"
29
[[2]]
[1] "c" "d"
For arrays, the names of the dimensions can be accessed with dimnames:
> A <- array(1:8, dim = c(2, 2, 2))
> A
, , 1
[1,]
[2,]
[,1] [,2]
1
3
2
4
, , 2
[1,]
[2,]
[,1] [,2]
5
7
6
8
> dimnames(A) <- list(c("a", "b"), c("c", "d"), c("e", "f"))
> A
, , e
c d
a 1 3
b 2 4
, , f
c d
a 5 7
b 6 8
If the elements of an object have names, they can be extracted by using
them as indices. Actually, this should be termed ‘subsetting’ rather than
‘extraction’ since the attributes of the original object are kept. For instance,
if a data frame DF contains the variables x, y, and z, the command DF["x"]
will return a data frame with just x; DF[c("x", "y")] will return a data
frame with both variables. This works with lists as well if the elements in the
list have names.
As the reader surely realizes, the index used here is a vector of mode
character. Like the numeric or logical vectors seen above, this vector can be
defined beforehand and then used for the extraction.
To extract a vector or a factor from a data frame, on can use the operator
$ (e.g., DF$x). This also works with lists.
30
3.5.6
The data editor
It is possible to use a graphical spreadsheet-like editor to edit a “data” object.
For example, if X is a matrix, the command data.entry(X) will open a graphic
editor and one will be able to modify some values by clicking on the appropriate
cells, or to add new columns or rows.
The function data.entry modifies directly the object given as argument
without needing to assign its result. On the other hand, the function de
returns a list with the objects given as arguments and possibly modified. This
result is displayed on the screen by default, but, as for most functions, can be
assigned to an object.
The details of using the data editor depend on the operating system.
3.5.7
Arithmetics and simple functions
There are numerous functions in R to manipulate data. We have already seen
the simplest one, c which concatenates the objects listed in parentheses. For
example:
> c(1:5, seq(10, 11, 0.2))
[1] 1.0 2.0 3.0 4.0 5.0 10.0 10.2 10.4 10.6 10.8 11.0
Vectors can be manipulated with classical arithmetic expressions:
> x
> y
> z
> z
[1]
<- 1:4
<- rep(1, 4)
<- x + y
2 3 4 5
Vectors of different lengths can be added; in this case, the shortest vector
is recycled. Examples:
> x <- 1:4
> y <- 1:2
> z <- x + y
> z
[1] 2 4 4 6
> x <- 1:3
> y <- 1:2
> z <- x + y
Warning message:
longer object length
is not a multiple of shorter object length in: x + y
> z
[1] 2 4 4
31
Note that R returned a warning message and not an error message, thus
the operation has been performed. If we want to add (or multiply) the same
value to all the elements of a vector:
> x
> a
> z
> z
[1]
<- 1:4
<- 10
<- a * x
10 20 30 40
The functions available in R for manipulating data are too many to be
listed here. One can find all the basic mathematical functions (log, exp,
log10, log2, sin, cos, tan, asin, acos, atan, abs, sqrt, . . . ), special functions (gamma, digamma, beta, besselI, . . . ), as well as diverse functions useful
in statistics. Some of these functions are listed in the following table.
sum(x)
prod(x)
max(x)
min(x)
which.max(x)
which.min(x)
range(x)
length(x)
mean(x)
median(x)
var(x) or cov(x)
cor(x)
var(x, y) or cov(x, y)
cor(x, y)
sum of the elements of x
product of the elements of x
maximum of the elements of x
minimum of the elements of x
returns the index of the greatest element of x
returns the index of the smallest element of x
id. than c(min(x), max(x))
number of elements in x
mean of the elements of x
median of the elements of x
variance of the elements of x (calculated on n − 1); if x is
a matrix or a data frame, the variance-covariance matrix is
calculated
correlation matrix of x if it is a matrix or a data frame (1 if x
is a vector)
covariance between x and y, or between the columns of x and
those of y if they are matrices or data frames
linear correlation between x and y, or correlation matrix if they
are matrices or data frames
These functions return a single value (thus a vector of length one), except
range which returns a vector of length two, and var, cov, and cor which may
return a matrix. The following functions return more complex results.
round(x, n)
rev(x)
sort(x)
rank(x)
rounds the elements of x to n decimals
reverses the elements of x
sorts the elements of x in increasing order; to sort in decreasing order:
rev(sort(x))
ranks of the elements of x
32
log(x, base)
scale(x)
pmin(x,y,...)
pmax(x,y,...)
cumsum(x)
cumprod(x)
cummin(x)
cummax(x)
match(x, y)
which(x == a)
choose(n, k)
na.omit(x)
na.fail(x)
unique(x)
table(x)
table(x, y)
subset(x, ...)
sample(x, size)
3.5.8
computes the logarithm of x with base base
if x is a matrix, centers and reduces the data; to center only use
the option center=FALSE, to reduce only scale=FALSE (by default
center=TRUE, scale=TRUE)
a vector which ith element is the minimum of x[i], y[i], . . .
id. for the maximum
a vector which ith element is the sum from x[1] to x[i]
id. for the product
id. for the minimum
id. for the maximum
returns a vector of the same length than x with the elements of x
which are in y (NA otherwise)
returns a vector of the indices of x if the comparison operation is
true (TRUE), in this example the values of i for which x[i] == a (the
argument of this function must be a variable of mode logical)
computes the combinations of k events among n repetitions = n!/[(n−
k)!k!]
suppresses the observations with missing data (NA) (suppresses the
corresponding line if x is a matrix or a data frame)
returns an error message if x contains at least one NA
if x is a vector or a data frame, returns a similar object but with the
duplicate elements suppressed
returns a table with the numbers of the differents values of x (typically
for integers or factors)
contingency table of x and y
returns a selection of x with respect to criteria (..., typically comparisons: x$V1 < 10); if x is a data frame, the option select gives
the variables to be kept (or dropped using a minus sign)
resample randomly and without replacement size elements in the
vector x, the option replace = TRUE allows to resample with replacement
Matrix computation
R has facilities for matrix computation and manipulation. The functions
rbind and cbind bind matrices with respect to the lines or the columns,
respectively:
> m1 <- matrix(1, nr = 2, nc = 2)
> m2 <- matrix(2, nr = 2, nc = 2)
> rbind(m1, m2)
[,1] [,2]
[1,]
1
1
[2,]
1
1
[3,]
2
2
[4,]
2
2
> cbind(m1, m2)
[,1] [,2] [,3] [,4]
33
[1,]
[2,]
1
1
1
1
2
2
2
2
The operator for the product of two matrices is ‘%*%’. For example, considering the two matrices m1 and m2 above:
> rbind(m1, m2) %*% cbind(m1, m2)
[,1] [,2] [,3] [,4]
[1,]
2
2
4
4
[2,]
2
2
4
4
[3,]
4
4
8
8
[4,]
4
4
8
8
> cbind(m1, m2) %*% rbind(m1, m2)
[,1] [,2]
[1,]
10
10
[2,]
10
10
The transposition of a matrix is done with the function t; this function
works also with a data frame.
The function diag can be used to extract or modify the diagonal of a
matrix, or to build a diagonal matrix.
> diag(m1)
[1] 1 1
> diag(rbind(m1, m2) %*% cbind(m1, m2))
[1] 2 2 8 8
> diag(m1) <- 10
> m1
[,1] [,2]
[1,]
10
1
[2,]
1
10
> diag(3)
[,1] [,2] [,3]
[1,]
1
0
0
[2,]
0
1
0
[3,]
0
0
1
> v <- c(10, 20, 30)
> diag(v)
[,1] [,2] [,3]
[1,]
10
0
0
[2,]
0
20
0
[3,]
0
0
30
> diag(2.1, nr = 3, nc = 5)
[,1] [,2] [,3] [,4] [,5]
[1,] 2.1 0.0 0.0
0
0
[2,] 0.0 2.1 0.0
0
0
[3,] 0.0 0.0 2.1
0
0
34
R has also some special functions for matrix computation. We can cite
here solve for inverting a matrix, qr for decomposition, eigen for computing
eigenvalues and eigenvectors, and svd for singular value decomposition.
35
4
Graphics with R
R offers a remarkable variety of graphics. To get an idea, one can type
demo(graphics) or demo(persp). It is not possible to detail here the possibilities of R in terms of graphics, particularly since each graphical function
has a large number of options making the production of graphics very flexible.
The way graphical functions work deviates substantially from the scheme
sketched at the beginning of this document. Particularly, the result of a graphical function cannot be assigned to an object 11 but is sent to a graphical device.
A graphical device is a graphical window or a file.
There are two kinds of graphical functions: the high-level plotting functions which create a new graph, and the low-level plotting functions which
add elements to an existing graph. The graphs are produced with respect to
graphical parameters which are defined by default and can be modified with
the function par.
We will see in a first time how to manage graphics and graphical devices;
we will then somehow detail the graphical functions and parameters. We will
see a practical example of the use of these functionalities in producing graphs.
Finally, we will see the packages grid and lattice whose functioning is different
from the one summarized above.
4.1
4.1.1
Managing graphics
Opening several graphical devices
When a graphical function is executed, if no graphical device is open, R opens a
graphical window and displays the graph. A graphical device may be open with
an appropriate function. The list of available graphical devices depends on the
operating system. The graphical windows are called X11 under Unix/Linux
and windows under Windows. In all cases, one can open a graphical window
with the command x11() which also works under Windows because of an alias
towards the command windows(). A graphical device which is a file will be
open with a function depending on the format: postscript(), pdf(), png(),
. . . The list of available graphical devices can be found with ?device.
The last open device becomes the active graphical device on which all
subsequent graphs are displayed. The function dev.list() displays the list
of open devices:
> x11(); x11(); pdf()
> dev.list()
11
There are a few remarkable exceptions: hist() and barplot() produce also numeric
results as lists or matrices.
36
X11 X11 pdf
2
3
4
The figures displayed are the device numbers which must be used to change
the active device. To know what is the active device:
> dev.cur()
pdf
4
and to change the active device:
> dev.set(3)
X11
3
The function dev.off() closes a device: by default the active device is
closed, otherwise this is the one which number is given as argument to the
function. R then displays the number of the new active device:
> dev.off(2)
X11
3
> dev.off()
pdf
4
Two specific features of the Windows version of R are worth mentioning:
a Windows Metafile device can be open with the function win.metafile, and
a menu “History” displayed when the graphical window is selected allowing
recording of all graphs drawn during a session (by default, the recording system
is off, the user switches it on by clicking on “Recording” in this menu).
4.1.2
Partitioning a graphic
The function split.screen partitions the active graphical device. For example:
> split.screen(c(1, 2))
divides the device into two parts which can be selected with screen(1) or
screen(2); erase.screen() deletes the last drawn graph. A part of the
device can itself be divided with split.screen() leading to the possibility to
make complex arrangements.
These functions are incompatible with others (such as layout or coplot)
and must not be used with multiple graphical devices. Their use should be
limited, for instance, to graphical exploration of data.
The function layout partitions the active graphic window in several parts
where the graphs will be displayed successively. Its main argument is a matrix with integer numbers indicating the numbers of the “sub-windows”. For
example, to divide the device into four equal parts:
37
> layout(matrix(1:4, 2, 2))
It is of course possible to create this matrix previously allowing to better
visualize how the device is divided:
> mat <- matrix(1:4, 2, 2)
> mat
[,1] [,2]
[1,]
1
3
[2,]
2
4
> layout(mat)
To actually visualize the partition created, one can use the function layout.show
with the number of sub-windows as argument (here 4). In this example, we
will have:
1
3
2
4
> layout.show(4)
The following examples show some of the possibilities offered by layout().
> layout(matrix(1:6, 3, 2))
> layout.show(6)
> layout(matrix(1:6, 2, 3))
> layout.show(6)
> m <- matrix(c(1:3, 3), 2, 2)
> layout(m)
> layout.show(3)
1
4
2
5
3
6
1
3
5
2
4
6
1
3
2
In all these examples, we have not used the option byrow of matrix(), the
sub-windows are thus numbered column-wise; one can just specify matrix(...,
byrow=TRUE) so that the sub-windows are numbered row-wise. The numbers
38
in the matrix may also be given in any order, for example, matrix(c(2, 1,
4, 3), 2, 2).
By default, layout() partitions the device with regular heights and widths:
this can be modified with the options widths and heights. These dimensions
are given relatively12 . Examples:
> m <- matrix(1:4, 2, 2)
> layout(m, widths=c(1, 3),
heights=c(3, 1))
> layout.show(4)
1
3
2
4
2
> m <- matrix(c(1,1,2,1),2,2)
> layout(m, widths=c(2, 1),
heights=c(1, 2))
> layout.show(2)
1
Finally, the numbers in the matrix can include zeros giving the possibility
to make complex (or even esoterical) partitions.
2
> m <- matrix(0:3, 2, 2)
> layout(m, c(1, 3), c(1, 3))
> layout.show(3)
1
> m <- matrix(scan(), 5, 5)
1: 0 0 3 3 3 1 1 3 3 3
11: 0 0 3 3 3 0 2 2 0 5
21: 4 2 2 0 5
26:
Read 25 items
> layout(m)
> layout.show(5)
12
3
4
1
2
3
5
They can be given in centimetres, see ?layout.
39
4.2
Graphical functions
Here is an overview of the high-level graphical functions in R.
plot(x)
plot(x, y)
sunflowerplot(x,
y)
pie(x)
boxplot(x)
stripchart(x)
coplot(x~y | z)
interaction.plot
(f1, f2, y)
matplot(x,y)
dotchart(x)
fourfoldplot(x)
assocplot(x)
mosaicplot(x)
pairs(x)
plot.ts(x)
ts.plot(x)
hist(x)
barplot(x)
qqnorm(x)
qqplot(x, y)
contour(x, y, z)
filled.contour (x,
y, z)
image(x, y, z)
persp(x, y, z)
stars(x)
plot of the values of x (on the y-axis) ordered on the x-axis
bivariate plot of x (on the x-axis) and y (on the y-axis)
id. but the points with similar coordinates are drawn as a flower
which petal number represents the number of points
circular pie-chart
“box-and-whiskers” plot
plot of the values of x on a line (an alternative to boxplot() for
small sample sizes)
bivariate plot of x and y for each value (or interval of values) of
z
if f1 and f2 are factors, plots the means of y (on the y-axis) with
respect to the values of f1 (on the x-axis) and of f2 (different
curves); the option fun allows to choose the summary statistic
of y (by default fun=mean)
bivariate plot of the first column of x vs. the first one of y, the
second one of x vs. the second one of y, etc.
if x is a data frame, plots a Cleveland dot plot (stacked plots
line-by-line and column-by-column)
visualizes, with quarters of circles, the association between two
dichotomous variables for different populations (x must be an
array with dim=c(2, 2, k), or a matrix with dim=c(2, 2) if
k = 1)
Cohen–Friendly graph showing the deviations from independence of rows and columns in a two dimensional contingency
table
‘mosaic’ graph of the residuals from a log-linear regression of a
contingency table
if x is a matrix or a data frame, draws all possible bivariate plots
between the columns of x
if x is an object of class "ts", plot of x with respect to time, x
may be multivariate but the series must have the same frequency
and dates
id. but if x is multivariate the series may have different dates
and must have the same frequency
histogram of the frequencies of x
histogram of the values of x
quantiles of x with respect to the values expected under a normal
law
quantiles of y with respect to the quantiles of x
contour plot (data are interpolated to draw the curves), x
and y must be vectors and z must be a matrix so that
dim(z)=c(length(x), length(y)) (x and y may be omitted)
id. but the areas between the contours are coloured, and a legend
of the colours is drawn as well
id. but the actual data are represented with colours
id. but in perspective
if x is a matrix or a data frame, draws a graph with segments
or a star where each row of x is represented by a star and the
columns are the lengths of the segments
40
symbols(x, y, ...)
termplot(mod.obj)
draws, at the coordinates given by x and y, symbols (circles,
squares, rectangles, stars, thermometres or “boxplots”) which
sizes, colours, etc, are specified by supplementary arguments
plot of the (partial) effects of a regression model (mod.obj)
For each function, the options may be found with the on-line help in R.
Some of these options are identical for several graphical functions; here are
the main ones (with their possible default values):
add=FALSE
axes=TRUE
type="p"
xlim=, ylim=
xlab=, ylab=
main=
sub=
4.3
if TRUE superposes the plot on the previous one (if it
exists)
if FALSE does not draw the axes and the box
specifies the type of plot, "p": points, "l": lines, "b":
points connected by lines, "o": id. but the lines are over
the points, "h": vertical lines, "s": steps, the data are
represented by the top of the vertical lines, "S": id. but
the data are represented by the bottom of the vertical
lines
specifies the lower and upper limits of the axes, for example with xlim=c(1, 10) or xlim=range(x)
annotates the axes, must be variables of mode character
main title, must be a variable of mode character
sub-title (written in a smaller font)
Low-level plotting commands
R has a set of graphical functions which affect an already existing graph: they
are called low-level plotting commands. Here are the main ones:
points(x, y)
lines(x, y)
text(x, y, labels,
...)
mtext(text,
side=3, line=0,
...)
segments(x0, y0,
x1, y1)
arrows(x0, y0,
x1, y1, angle= 30,
code=2)
abline(a,b)
abline(h=y)
abline(v=x)
abline(lm.obj)
adds points (the option type= can be used)
id. but with lines
adds text given by labels at coordinates (x,y); a typical use is:
plot(x, y, type="n"); text(x, y, names)
adds text given by text in the margin specified by side (see
axis() below); line specifies the line from the plotting area
draws lines from points (x0,y0) to points (x1,y1)
id. with arrows at points (x0,y0) if code=2, at points (x1,y1) if
code=1, or both if code=3; angle controls the angle from the
shaft of the arrow to the edge of the arrow head
draws a line of slope b and intercept a
draws a horizontal line at ordinate y
draws a vertical line at abcissa x
draws the regression line given by lm.obj (see section 5)
41
rect(x1, y1, x2,
y2)
polygon(x, y)
legend(x, y,
legend)
title()
axis(side, vect)
box()
rug(x)
locator(n,
type="n", ...)
draws a rectangle which left, right, bottom, and top limits are
x1, x2, y1, and y2, respectively
draws a polygon linking the points with coordinates given by x
and y
adds the legend at the point (x,y) with the symbols given by
legend
adds a title and optionally a sub-title
adds an axis at the bottom (side=1), on the left (2), at the top
(3), or on the right (4); vect (optional) gives the abcissa (or
ordinates) where tick-marks are drawn
adds a box around the current plot
draws the data x on the x-axis as small vertical lines
returns the coordinates (x, y) after the user has clicked n times
on the plot with the mouse; also draws symbols (type="p") or
lines (type="l") with respect to optional graphic parameters
(...); by default nothing is drawn (type="n")
Note the possibility to add mathematical expressions on a plot with text(x,
y, expression(...)), where the function expression transforms its argument in a mathematical equation. For example,
> text(x, y, expression(p == over(1, 1+e^-(beta*x+alpha))))
will display, on the plot, the following equation at the point of coordinates
(x, y):
1
1 + e−(β x+α)
To include in an expression a variable we can use the functions substitute
and as.expression; for example to include a value of R 2 (previously computed and stored in an object named Rsquared):
p=
> text(x, y, as.expression(substitute(R^2==r, list(r=Rsquared))))
will display on the plot at the point of coordinates (x, y):
R2 = 0.9856298
To display only three decimals, we can modify the code as follows:
> text(x, y, as.expression(substitute(R^2==r,
+
list(r=round(Rsquared, 3)))))
will display:
R2 = 0.986
Finally, to write the R in italics:
> text(x, y, as.expression(substitute(italic(R)^2==r,
+
list(r=round(Rsquared, 3)))))
R2 = 0.986
42
4.4
Graphical parameters
In addition to low-level plotting commands, the presentation of graphics can
be improved with graphical parameters. They can be used either as options
of graphic functions (but it does not work for all), or with the function par to
change permanently the graphical parameters, i.e. the subsequent plots will
be drawn with respect to the parameters specified by the user. For instance,
the following command:
> par(bg="yellow")
will result in all subsequent plots drawn with a yellow background. There
are 73 graphical parameters, some of them have very similar functions. The
exhaustive list of these parameters can be read with ?par; I will limit the
following table to the most usual ones.
adj
bg
bty
cex
col
font
las
lty
lwd
mar
mfcol
mfrow
pch
ps
controls text justification with respect to the left border of the text so that
0 is left-justified, 0.5 is centred, 1 is right-justified, values > 1 move the text
further to the left, and negative values further to the right; if two values are
given (e.g., c(0, 0)) the second one controls vertical justification with respect
to the text baseline
specifies the colour of the background (e.g., bg="red", bg="blue"; the list of
the 657 available colours is displayed with colors())
controls the type of box drawn around the plot, allowed values are: "o",
"l", "7", "c", "u" ou "]" (the box looks like the corresponding character); if
bty="n" the box is not drawn
a value controlling the size of texts and symbols with respect to the default; the
following parameters have the same control for numbers on the axes, cex.axis,
the axis labels, cex.lab, the title, cex.main, and the sub-title, cex.sub
controls the colour of symbols; as for cex there are: col.axis, col.lab,
col.main, col.sub
an integer which controls the style of text (1: normal, 2: italics, 3: bold, 4:
bold italics); as for cex there are: font.axis, font.lab, font.main, font.sub
an integer which controls the orientation of the axis labels (0: parallel to the
axes, 1: horizontal, 2: perpendicular to the axes, 3: vertical)
controls the type of lines, can be an integer (1: solid, 2: dashed, 3: dotted,
4: dotdash, 5: longdash, 6: twodash), or a string of up to eight characters
(between "0" and "9") which specifies alternatively the length, in points or
pixels, of the drawn elements and the blanks, for example lty="44" will have
the same effet than lty=2
a numeric which controls the width of lines
a vector of 4 numeric values which control the space between the axes and the
border of the graph of the form c(bottom, left, top, right), the default
values are c(5.1, 4.1, 4.1, 2.1)
a vector of the form c(nr,nc) which partitions the graphic window as a matrix of nr lines and nc columns, the plots are then drawn in columns (see
section 4.1.2)
id. but the plots are then drawn in line (see section 4.1.2)
controls the type of symbol, either an integer between 1 and 25, or any single
character within "" (Fig. 2)
an integer which controls the size in points of texts and symbols
43
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
"*"
"?"
"."
"X"
"a"
*
?
X a
Figure 2: The plotting symbols in R (pch=1:25). The colours were obtained
with the options col="blue", bg="yellow", the second option has an effect
only for the symbols 21–25. Any character can be used (pch="*", "?", ".",
. . . ).
pty
tck
tcl
xaxt
yaxt
4.5
a character which specifies the type of the plotting region, "s": square, "m":
maximal
a value which specifies the length of tick-marks on the axes as a fraction of
the smallest of the width or height of the plot; if tck=1 a grid is drawn
id. but as a fraction of the height of a line of text (by default tcl=-0.5)
if xaxt="n" the x-axis is set but not drawn (useful in conjunction with
axis(side=1, ...))
if yaxt="n" the y-axis is set but not drawn (useful in conjunction with
axis(side=2, ...))
A practical example
In order to illustrate R’s graphical functionalities, let us consider a simple
example of a bivariate graph of 10 pairs of random variates. These values
were generated with:
> x <- rnorm(10)
> y <- rnorm(10)
The wanted graph will be obtained with plot(); one will type the command:
> plot(x, y)
and the graph will be plotted on the active graphical device. The result
is shown on Fig. 3. By default, R makes graphs in an “intelligent” way:
44
0.5
0.0
−1.0
−0.5
y
−0.5
0.0
0.5
1.0
x
Figure 3: The function plot used without options.
the spaces between tick-marks on the axes, the placement of labels, etc, are
calculated so that the resulting graph is as intelligible as possible.
The user may, nevertheless, change the way a graph is presented, for instance, to conform to a pre-defined editorial style, or to give it a personal
touch for a talk. The simplest way to change the presentation of a graph is to
add options which will modify the default arguments. In our example, we can
modify significantly the figure in the following way:
plot(x, y, xlab="Ten random values", ylab="Ten other values",
xlim=c(-2, 2), ylim=c(-2, 2), pch=22, col="red",
bg="yellow", bty="l", tcl=0.4,
main="How to customize a plot with R", las=1, cex=1.5)
The result is on Fig. 4. Let us detail each of the used options. First,
xlab and ylab change the axis labels which, by default, were the names of the
variables. Then, xlim and ylim allow us to define the limits on both axes 13 .
The graphical parameter pch is used here as an option: pch=22 specifies a
square which contour and background colours may be different and are given
by, respectively, col and bg. The table of graphical parameters gives the
meaning of the modifications done by bty, tcl, las and cex. Finally, a title
is added with the option main.
The graphical parameters and the low-level plotting functions allow us to
go further in the presentation of a graph. As we have seen previously, some
graphical parameters cannot be passed as arguments to a function like plot.
13
By default, R adds 4% on each side of the axis limit. This behaviour may be altered by
setting the graphical parameters xaxs="i" and yaxs="i" (they can be passed as options to
plot()).
45
How to customize a plot with R
2
Ten other values
1
0
−1
−2
−2
−1
0
1
2
Ten random values
Figure 4: The function plot used with options.
We will now modify some of these parameters with par(), it is thus necessary
to type several commands. When the graphical parameters are changed, it
is useful to save their initial values beforehand to be able to restore them
afterwards. Here are the commands used to obtain Fig. 5.
opar <- par()
par(bg="lightyellow", col.axis="blue", mar=c(4, 4, 2.5, 0.25))
plot(x, y, xlab="Ten random values", ylab="Ten other values",
xlim=c(-2, 2), ylim=c(-2, 2), pch=22, col="red", bg="yellow",
bty="l", tcl=-.25, las=1, cex=1.5)
title("How to customize a plot with R (bis)", font.main=3, adj=1)
par(opar)
Let us detail the actions resulting from these commands. First, the default
graphical parameters are copied in a list called here opar. Three parameters
will be then modified: bg for the colour of the background, col.axis for the
colour of the numbers on the axes, and mar for the sizes of the margins around
the plotting region. The graph is drawn in a nearly similar way to Fig. 4. The
modifications of the margins allowed to use the space around the plotting area.
The title here is added with the low-level plotting function title which allows
to give some parameters as arguments without altering the rest of the graph.
Finally, the initial graphical parameters are restored with the last command.
Now, total control! On Fig. 5, R still determines a few things such as
the number of tick marks on the axes, or the space between the title and the
plotting area. We will see now how to totally control the presentation of the
graph. The approach used here is to plot a “blank” graph with plot(...,
type="n"), then to add points, axes, labels, etc, with low-level plotting func46
How to customize a plot with R (bis)
2
Ten other values
1
0
−1
−2
−2
−1
0
1
2
Ten random values
Figure 5: The functions par, plot and title.
tions. We will fancy a few arrangements such as changing the colour of the
plotting area. The commands follow, and the resulting graph is on Fig. 6.
opar <- par()
par(bg="lightgray", mar=c(2.5, 1.5, 2.5, 0.25))
plot(x, y, type="n", xlab="", ylab="", xlim=c(-2, 2),
ylim=c(-2, 2), xaxt="n", yaxt="n")
rect(-3, -3, 3, 3, col="cornsilk")
points(x, y, pch=10, col="red", cex=2)
axis(side=1, c(-2, 0, 2), tcl=-0.2, labels=FALSE)
axis(side=2, -1:1, tcl=-0.2, labels=FALSE)
title("How to customize a plot with R (ter)",
font.main=4, adj=1, cex.main=1)
mtext("Ten random values", side=1, line=1, at=1, cex=0.9, font=3)
mtext("Ten other values", line=0.5, at=-1.8, cex=0.9, font=3)
mtext(c(-2, 0, 2), side=1, las=1, at=c(-2, 0, 2), line=0.3,
col="blue", cex=0.9)
mtext(-1:1, side=2, las=1, at=-1:1, line=0.2, col="blue", cex=0.9)
par(opar)
Like before, the default graphical parameters are saved, and the colour
of the background and the margins are modified. The graph is then drawn
with type="n" to not plot the points, xlab="", ylab="" to not write the
axis labels, and xaxt="n", yaxt="n" to not draw the axes. This results in
drawing only the box around the plotting area, and defining the axes with
respect to xlim et ylim. Note that we could have used the option axes=FALSE
but in this case neither the axes, nor the box would have been drawn.
47
Ten other values
How to customize a plot with R (ter)
1
0
−1
−2
0
Ten random values
2
Figure 6: A “hand-made” graph.
The elements are then added in the plotting region so defined with some
low-level plotting functions. Before adding the points, the colour inside the
plotting area is changed with rect(): the size of the rectangle are chosen so
that it is substantially larger than the plotting area.
The points are plotted with points(); a new symbol was used. The axes
are added with axis(): the vector given as second argument specifies the
coordinates of the tick-marks. The option labels=FALSE specifies that no
annotation must be written with the tick-marks. This option also accepts a
vector of mode character, for example labels=c("A", "B", "C").
The title is added with title(), but the font is slightly changed. The
annotations on the axes are written with mtext() (marginal text). The first
argument of this function is a vector of mode character giving the text to be
written. The option line indicates the distance from the plotting area (by
default line=0), and at the coordinnate. The second call to mtext() uses
the default value of side (3). The two other calls to mtext() pass a numeric
vector as first argument: this will be converted into character.
4.6
The grid and lattice packages
The packages grid and lattice implement the grid and lattice systems. Grid is
a new graphical mode with its own system of graphical parameters which are
distinct from those seen above. The two main distinctions of grid compared
to the base graphics are:
• a more flexible way to split graphical devices using viewports which could
be overpalling (graphical objects may even be shared among distinct
viewports, e.g., arrows);
48
• graphical objects (grob) may be modified or removed from a graph without requiring the re-draw all the graph (as must be done with base
graphics).
Grid graphics cannot usually be combined or mixed with base graphics
(the gridBase package must be used to do this). However, it is possible to use
both graphical modes in the same session on the same graphical device.
Lattice is essentially the implementation in R of the Trellis graphics of
S-PLUS. Trellis is an approach for visualizing multivariate data which is particularly appropriate for the exploration of relations or interactions among
variables14 . The main idea behind lattice (and Trellis as well) is that of conditional multiple graphs: a bivariate graph will be split in several graphs with
respect to the values of a third variable. The function coplot uses a similar
approach, but lattice offers much wider functionalities. Lattice uses the grid
graphical mode.
Most functions in lattice take a formula as their main argument 15 , for
example y ~ x. The formula y ~ x | z means that the graph of y with
respect to x will be plotted as several graphs with respect to the values of z.
The following table gives the main functions in lattice. The formula given
as argument is the typical necessary formula, but all these functions accept
a conditional formula (y ~ x | z) as main argument; in the latter case, a
multiple graph, with respect to the values of z, is plotted as will be seen in
the examples below.
barchart(y ~ x)
bwplot(y ~ x)
densityplot(~ x)
dotplot(y ~ x)
histogram(~ x)
qqmath(~ x)
stripplot(y ~ x)
qq(y ~ x)
xyplot(y ~ x)
levelplot(z ~ x*y)
contourplot(z ~ x*y)
cloud(z ~ x*y)
wireframe(z ~ x*y)
splom(~ x)
parallel(~ x)
histogram of the values of y with respect to those of x
“box-and-whiskers” plot
density functions plot
Cleveland dot plot (stacked plots line-by-line and
column-by-column)
histogram of the frequencies of x
quantiles of x with respect to the values expected under
a theoretical distribution
single dimension plot, x must be numeric, y may be a
factor
quantiles to compare two distributions, x must be numeric, y may be numeric, character, or factor but must
have two ‘levels’
bivariate plots (with many functionalities)
coloured plot of the values of z at the coordinates given
by x and y (x, y and z are all of the same length)
3-D perspective plot (points)
id. (surface)
matrix of bivariate plots
parallel coordinates plot
Let us see now some examples in order to illustrate a few aspects of lattice.
The package must be loaded in memory with the command library(lattice)
14
http://cm.bell-labs.com/cm/ms/departments/sia/project/trellis/index.html
plot() also accepts a formula as its main argument: if x and y are two vectors of the
same length, plot(y ~ x) and plot(x, y) will give identical graphs.
15
49
−4
−2
0
2
4
n = 35
n = 40
n = 45
n = 20
n = 25
n = 30
0.6
0.5
0.4
0.3
0.2
0.1
0
0.6
Density
0.5
0.4
0.3
0.2
0.1
0
n=5
n = 10
n = 15
0.6
0.5
0.4
0.3
0.2
0.1
0
−4
−2
0
2
4
−4
−2
0
2
4
x
Figure 7: The function densityplot.
so that the functions can be accessed.
Let us start with the graphs of density functions. Such graphs can be
done simply with densityplot(~ x) which will plot a curve of the empirical
density function with the points corresponding to the observations on the xaxis (similarly to rug()). Our example will be slightly more complicated with
the superposition, on each plot, of the curves of empirical density and those
predicted from a normal law. It is necessary to use the argument panel which
defines what is drawn on each plot. The commands are:
n <- seq(5, 45, 5)
x <- rnorm(sum(n))
y <- factor(rep(n, n), labels=paste("n =", n))
densityplot(~ x | y,
panel = function(x, ...) {
panel.densityplot(x, col="DarkOliveGreen", ...)
panel.mathdensity(dmath=dnorm,
args=list(mean=mean(x), sd=sd(x)),
col="darkblue")
})
The first three lines of command generate a random sample of independent
normal variates which is split in sub-samples of size equal to 5, 10, 15, . . . , and
45. Then comes the call to densityplot producing a plot for each sub-sample.
panel takes as argument a function. In our example, we have defined a function which calls two functions pre-defined in lattice: panel.densityplot to
draw the empirical density function, and panel.mathdensity to draw the density function predicted from a normal law. The function panel.densityplot
is called by default if no argument is given to panel: the command densityplot
50
165 170 175
180 185
472−544
544−616
616−688
256−328
328−400
400−472
−10
−15
−20
−25
−30
−35
−40
−10
−15
lat
−20
−25
−30
−35
−40
40−112
112−184
184−256
−10
−15
−20
−25
−30
−35
−40
165 170 175 180
185
165
170 175 180 185
long
Figure 8: The function xyplot with the data “quakes”.
(~ x | y) would have resulted in the same graph than Fig. 7 but without the
blue curves.
The next examples are taken, more or less modified, from the help pages
of lattice, and use some data sets available in R: the locations of 1000 seismic
events near the Fiji Islands, and some flower measurements made on three
species of iris.
Fig. 8 shows the geographic locations of the seismic events with respect to
depth. The commands necessary for this graph are:
data(quakes)
mini <- min(quakes$depth)
maxi <- max(quakes$depth)
int <- ceiling((maxi - mini)/9)
inf <- seq(mini, maxi, int)
quakes$depth.cat <- factor(floor(((quakes$depth - mini) / int)),
labels=paste(inf, inf + int, sep="-"))
xyplot(lat ~ long | depth.cat, data = quakes)
The first command loads the data quakes in memory. The five next commands create a factor by dividing the depth (variable depth) in nine equallyranged intervals: the levels of this factor are labelled with the lower and upper
bounds of these intervals. It then suffices to call the function xyplot with the
appropriate formula and an argument data indicating where xyplot must look
for the variables16 .
With the data iris, the overlap among the different species is sufficiently
small so they can be plotted on the figure (Fig. 9). The commands are:
16
plot() cannot take an argument data, the location of the variables must be given
explicitly, for example plot(quakes$long ~ quakes$lat).
51
7
o
o
o
setosa
versicolor
virginica
6
o
o
o
o
o
o
5
Petal.Length
o
o
o
4
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
3
2
o
o
o
1
0
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
0.5
o
1
1.5
2
2.5
Petal.Width
Figure 9: The function xyplot with the data “iris”.
data(iris)
xyplot(
Petal.Length ~ Petal.Width, data = iris, groups=Species,
panel = panel.superpose,
type = c("p", "smooth"), span=.75,
auto.key = list(x = 0.15, y = 0.85)
)
The call to the function xyplot is here a bit more complex than in the
previous example and uses several options that we will detail. The option
groups, as suggested by its name, defines groups that will be used by the
other options. We have already seen the option panel which defines how the
different groups will be represented on the graph: we use here a pre-defined
function panel.superpose in order to superpose the groups on the same plot.
No option is passed to panel.superpose, the default colours will be used to
distinguish the groups. The option type, like in plot(), specifies how the
data are represented, but here we can give several arguments as a vector:
"p" to draw points and "smooth" to draw a smooth curve which degree of
smoothness is specified by span. The option auto.key adds a legend to the
graph: it is only necessary to give, as a list, the coordinates where the legend
is to be plotted. Note that here these coordinates are relative to the size of
the plot (i.e. in [0, 1]).
We will see now the function splom with the same data on iris. The
following commands were used to produce Fig. 10:
splom(
~iris[1:4], groups = Species, data = iris, xlab = "",
panel = panel.superpose,
52
Setosa
Versicolor
o o
o o
o oo
ooo o o
ooo
oo
oo
oo o
o o
o
o
o
oooooo oo
oo
o
ooo oo
o
o
o
o
oooo
oo
ooo
oo
ooo
oooooooo
oo
o
o
o
o
o
ooo
oooo
ooooo
ooooo o
o
o
o
o
o
o
oo
oo
o
o ooo o
oo
oo
oo
oo
o
o
o
o
o
o
o
o
o
o
oo o
oo o
ooo
ooo
oo
ooo o
oo
o ooo
Virginica
oo
oooooo
o
ooooo oo
ooo
ooooooo ooo
ooo
oooooooo
o
o
ooo
o o
oooo
ooo
o o
ooo
ooo
oo
oo
oo
ooo
o
oooooo
oo
oo
o
o
o
ooo
ooo
oo
oooo
o
oooooo
oo
ooo o
2.5
1.5
7
Sepal.Length
6
5
5
6
o
oo oo
oo
oooo
oo
ooo
o o oo o 1
o o o oo
o
4.5
3.5
4
4
3.5
Sepal.Width
2
2
4.5 o
o
oooo
oooo
o
ooo
oo
oo
o
ooo
ooo
ooo
oo
3 ooo
oo
2.5
o
2
o
o
ooo
o
oo
oo o
o
oo
o
o
o ooo
o
oooo
oo
oooo
oo
oo
o
oo oooo
o
o o
o
ooo
oo
o
oo
o
o
o
ooo
o o o ooo
oooo
ooo
oo oo
o
oo o
o oo o
ooo
oo
o
oo
o
o
o
o
oo
o
o
oo
o ooo
o oo
oo
oooo
ooo
o
ooo
o
o
oo
oo o o
oo
oo
oo o o
oo
ooo
o
oo
o
o
o
oo
2
2.5
3
3
4
1
2
2.5
Petal.Width
1
oo
ooo ooo
o o
oo
o
o
oo
oooo
o
oo
oooo
oo
oo
ooo
oo
ooooo
0
oo
o ooo
oo
oooo
7
o
oo
4 5 6 7
o
oo
o
o
oo oo
ooo oo o
oooo oo
6
o
o
o
o
oooooooo
o oooo
oooo
o
o o
oo ooo
5
o
ooo
oooooo
o oo
oooo
oo o
oo
oooooooo
ooo
oo oo oo
oo
o oooo ooo
ooo o
ooooooooooo
oo
4 Petal.Length 4
oooooo
oo
o
oo
oo
oo
o
o
o
o
oo
oo
3
o
o
oo
ooo
oooo
ooooooooo
oooo o o o
o
ooo
oo
ooo
oo
o
o
o oo
oo
ooo ooo
ooo
oo
oo
ooo
ooo
oo
ooooo
o oo
oo
oo
oo
o
oo
o
ooooo oo
oo
oo oo
o o
o
o
o
o
o
o
o
o
o
oo
oo o o o
o
ooo
o oo
oo
oo
o
oo
o oo
oo o
o
ooo
o oo o
ooo
o
8
7
8
1.5
2
0.5
0.5
o
ooo
o
ooo
oo
ooo
o
oo
o
o
oo
oo
ooo
o
oo
o
oo
oo o
o
o
o o
o
oo
oo
o
oo
o
oo oooo
oooo
oooooo
oooo
ooooooo
ooooooo
oo
o
o
o oo
oooo
o
oo
ooo o o
ooooooooo
ooooooo
o o oo oo o
o
o o
o oo o
o
o
o ooo
o
oooo
oooooo
o oooooo
o
oooooo
ooo
oooooooo
oo
oo
o
o
o
o
ooo o
o
oooooo
ooo
oooooo
o
ooo
ooo
o
ooooooo
o
o
oo
oo
oooo
ooo o
o
ooo o
o o
o
oo
o
o
o
o
oo
oo
1
0
oooo
ooo o o
o
ooo o
o o
ooo
o
ooo
o
o
o
ooo oo
oooo
oooo
ooo
ooooo
o oo
oo o
oo oo
o
o
o
oo
o
ooo
o
o o oo o
oo
o oo
ooo
oo
ooo
oo
o
o
oo
o
o
o
o
oo
oo
oooo o
ooo o
oo
oo
o
ooooo oooo o
o
o
o
o oo
o
o
o
oooo
oo
oo
o o
oo
oo
o
ooo
oo
oo oo o
oo
o
o
o
o
oo
oo o
oo
o
o ooo
o
o o
ooo
o
o
o oooo
oo o
oo
o
o o oo
o o
ooo
oo
oo
oo
o
o
o
o
o
Figure 10: The function splom with the data “iris” (1).
auto.key = list(columns = 3)
)
The main argument is this time a matrix (the four first columns of iris).
The result is the set of possible bivariate plots among the columns of the
matrix, like the standard function pairs. By default, splom adds the text
“Scatter Plot Matrix” under the x-axis: to avoid this, the option xlab=""
was used. The other options are similar to the previous example, except that
columns = 3 for auto.key was specified so the legend is displayed in three
columns.
Fig. 10 could have been done with pairs(), but this latter function cannot
make conditional graphs like on Fig. 11. The code used is relatively simple:
splom(~iris[1:3] | Species, data = iris, pscales = 0,
varnames = c("Sepal\nLength", "Sepal\nWidth", "Petal\nLength"))
The sub-graphs being relatively small, we added two options to improve
the legibility of the figure: pscales = 0 removes the tick-marks on the axes
(all sub-graphs are drawn on the same scales), and the names of the variables
were re-defined to display them on two lines ("\n" codes for a line break in a
character string).
The last example uses the method of parallel coordinates for the exploratory analysis of multivariate data. The variables are arranged on an
axis (e.g., the y-axis), and the observed values are plotted on the other axis
(the variables are scaled similarly, e.g., by standardizing them). The different
values of the same individual are joined by a line. With the data iris, Fig. 12
is obtained with the following code:
parallel(~iris[, 1:4] | Species, data = iris, layout = c(3, 1))
53
virginica
Petal
Length
Sepal
Width
Sepal
Length
setosa
versicolor
Petal
Length
Petal
Length
Sepal
Width
Sepal
Width
Sepal
Length
Sepal
Length
Scatter Plot Matrix
Figure 11: The function splom with the data “iris” (2).
Min
setosa
Max
versicolor
virginica
Petal.Width
Petal.Length
Sepal.Width
Sepal.Length
Min
Max
Min
Max
Figure 12: The function parallel with the data “iris”.
54
5
Statistical analyses with R
Even more than for graphics, it is impossible here to go in the details of the
possibilities offered by R with respect to statistical analyses. My goal here is
to give some landmarks with the aim to have an idea of the features of R to
perform data analyses.
The package stats contains functions for a wide range of basic statistical analyses: classical tests, linear models (including least-squares regression,
generalized linear models, and analysis of variance), distributions, summary
statistics, hierarchical clustering, time-series analysis, nonlinear least squares,
and multivariate analysis. Other statistical methods are available in a large
number of packages. Some of them are distributed with a base installation
of R and are labelled recommanded, and many other packages are contributed
and must be installed by the user.
We will start with a simple example which requires no other package than
stats in order to introduce the general approach to data analysis in R. Then,
we will detail some notions, formulae and generic functions, which are useful
whatever the type of analysis performed. We will conclude with an overview
on packages.
5.1
A simple example of analysis of variance
The function for the analysis of variance in stats is aov. In order to try it, let
us take a data set distributed with R: InsectSprays. Six insecticides were
tested in field conditions, the observed response was the number of insects.
Each insecticide was tested 12 times, thus there are 72 observations. We will
not consider here the graphical exploration of the data, but will focus on a
simple analysis of variance of the response with respect to the insecticide. After
loading the data in memory with the function data, the analysis is performed
after a square-root transformation of the response:
> data(InsectSprays)
> aov.spray <- aov(sqrt(count) ~ spray, data = InsectSprays)
The main (and mandatory) argument of aov is a formula which specifies
the response on the left-hand side of the tilde symbol ~ and the predictor
on the right-hand side. The option data = InsectSprays specifies that the
variables must be found in the data frame InsectSprays. This syntax is
equivalent to:
> aov.spray <- aov(sqrt(InsectSprays$count) ~ InsectSprays$spray)
or still (if we know the column numbers of the variables):
55
> aov.spray <- aov(sqrt(InsectSprays[, 1]) ~ InsectSprays[, 2])
The first syntax is to be preferred since it is clearer.
The results are not displayed since they are assigned to an object called
aov.spray. We will then used some functions to extract the results, for example print to display a brief summary of the analysis (mostly the estimated parameters) and summary to display more details (included the statistical tests):
> aov.spray
Call:
aov(formula = sqrt(count) ~ spray, data = InsectSprays)
Terms:
spray Residuals
Sum of Squares 88.43787 26.05798
Deg. of Freedom
5
66
Residual standard error: 0.6283453
Estimated effects may be unbalanced
> summary(aov.spray)
Df Sum Sq Mean Sq F value
Pr(>F)
spray
5 88.438 17.688 44.799 < 2.2e-16 ***
Residuals
66 26.058
0.395
--Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
We may remind that typing the name of the object as a command is similar
to the command print(aov.spray). A graphical representation of the results
can be done with plot() or termplot(). Before typing plot(aov.spray) we
will divide the graphics into four parts so that the four diagnostics plots will
be done on the same graph. The commands are:
>
>
>
>
>
opar <- par()
par(mfcol = c(2, 2))
plot(aov.spray)
par(opar)
termplot(aov.spray, se=TRUE, partial.resid=TRUE, rug=TRUE)
and the resulting graphics are on Figs. 13 and 14.
5.2
Formulae
Formulae are a key element in statistical analyses with R: the notation used
is the same for (almost) all functions. A formula is typically of the form y
~ model where y is the analysed response and model is a set of terms for
which some parameters are to be estimated. These terms are separated with
arithmetic symbols but they have here a particular meaning.
56
1.5
2.5
0.5
3.5
1.5
1
2
0.08
27
25
0
Theoretical Quantiles
39
0.04
Cook’s distance
Cook’s distance plot
25
0
3.5
0.00
0
1
2
39 27
−1
2.5
Fitted values
Normal Q−Q plot
−2
Standardized residuals
Fitted values
−2
39
1.0
1.5
Scale−Location plot
27
25
0.0
25
Standardized residuals
1.0
39
0.0
−1.5
Residuals
Residuals vs Fitted
27
20
40
60
Obs. number
Figure 13: Graphical representation of the results from the function aov with
plot().
a+b
X
a:b
a*b
poly(a, n)
^n
b %in% a
-b
-1
1
offset(...)
additive effects of a and of b
if X is a matrix, this specifies an additive effect of each of
its columns, i.e. X[,1]+X[,2]+...+X[,ncol(X)]; some
of the columns may be selected with numeric indices (e.g.,
X[,2:4])
interactive effect between a and b
additive and interactive effects (identical to a+b+a:b)
polynomials of a up to degree n
includes all interactions up to level n, i.e. (a+b+c)^2 is
identical to a+b+c+a:b+a:c+b:c
the effects of b are nested in a (identical to a+a:b, or
a/b)
removes the effect of b, for example: (a+b+c)^2-a:b is
identical to a+b+c+a:c+b:c
y~x-1 is a regression through the origin (id. for y~x+0 or
0+y~x)
y~1 fits a model with no effects (only the intercept)
adds an effect to the model without estimating any parameter (e.g., offset(3*x))
We see that the arithmetic operators of R have in a formula a different
meaning than they have in expressions. For example, the formula y~x1+x2
defines the model y = β1 x1 + β2 x2 + α, and not (if the operator + would have
its usual meaning) y = β(x1 + x2 ) + α. To include arithmetic operations in a
formula, we can use the function I: the formula y~I(x1+x2) defines the model
y = β(x1 + x2 ) + α. Similarly, to define the model y = β 1 x + β2 x2 + α, we
will use the formula y ~ poly(x, 2) (and not y ~ x + x^2). However, it is
57
2
1
0
−1
−3
−2
Partial for spray
0
1
2
3
4
5
6
spray
Figure 14: Graphical representation of the results from the function aov with
termplot().
possible to include a function in a formula in order to transform a variable as
seen above with the insect sprays analysis of variance.
For analyses of variance, aov() accepts a particular syntax to define random effects. For instance, y ~ a + Error(b) means the additive effects of
the fixed term a and the random one b.
5.3
Generic functions
We remember that R’s functions act with respect to the attributes of the
objects possibly passed as arguments. The class is an attribute deserving some
attention here. It is very common that the R statistical functions return an
object of class with the same name (e.g., aov returns an object of class "aov",
lm returns one of class "lm"). The functions that we can use subsequently to
extract the results will act specifically with respect to the class of the object.
These functions are called generic.
For instance, the function which is most often used to extract results from
analyses is summary which displays detailed results. Whether the object given
as argument is of class "lm" (linear model) or "aov" (analysis of variance),
it sounds obvious that the information to display will not be the same. The
advantage of generic functions is that the syntax is the same in all cases.
An object containing the results of an analysis is generally a list, and the
way it is displayed is determined by its class. We have already seen this notion
that the action of a function depends on the kind of object given as argument.
It is a general feature of R17 . The following table gives the main generic func17
There are more than 100 generic functions in R.
58
tions which can be used to extract information from objects resulting from an
analysis. The typical usage of these functions is:
> mod <- lm(y ~ x)
> df.residual(mod)
[1] 8
print
summary
df.residual
coef
residuals
deviance
fitted
logLik
AIC
returns a brief summary
returns a detailed summary
returns the number of residual degrees of freedom
returns the estimated coefficients (sometimes with their standard-errors)
returns the residuals
returns the deviance
returns the fitted values
computes the logarithm of the likelihood and the number of parameters
computes the Akaike information criterion or AIC (depends on logLik())
A function like aov or lm returns a list with its different elements corresponding to the results of the analysis. If we take our example of an analysis
of variance with the data InsectSprays, we can look at the structure of the
object returned by aov:
> str(aov.spray, max.level = -1)
List of 13
- attr(*, "class")= chr [1:2] "aov" "lm"
Another way to look at this structure is to display the names of the object:
> names(aov.spray)
[1] "coefficients"
[4] "rank"
[7] "qr"
[10] "xlevels"
[13] "model"
"residuals"
"fitted.values"
"df.residual"
"call"
"effects"
"assign"
"contrasts"
"terms"
The elements can then be extracted as we have already seen:
> aov.spray$coefficients
(Intercept)
sprayB
sprayC
3.7606784
0.1159530 -2.5158217
sprayE
sprayF
-1.9512174
0.2579388
sprayD
-1.5963245
summary() also creates a list which, in the case of aov(), is simply a table
of tests:
59
> str(summary(aov.spray))
List of 1
$ :Classes anova and ‘data.frame’:
2 obs. of 5 variables:
..$ Df
: num [1:2] 5 66
..$ Sum Sq : num [1:2] 88.4 26.1
..$ Mean Sq: num [1:2] 17.688 0.395
..$ F value: num [1:2] 44.8
NA
..$ Pr(>F) : num [1:2] 0 NA
- attr(*, "class")= chr [1:2] "summary.aov" "listof"
> names(summary(aov.spray))
NULL
Generic functions do not generally perform any action on objects: they
call the appropriate function with respect to the class of the argument. A
function called by a generic is a method in R’s jargon. Schematically, a method
is constructed as generic.cls , where cls is the class of the object. For
instance, in the case of summary, we can display the corresponding methods:
> apropos("^summary")
[1] "summary"
[3] "summary.aovlist"
[5] "summary.data.frame"
[7] "summary.factor"
[9] "summary.glm.null"
[11] "summary.lm"
[13] "summary.manova"
[15] "summary.mlm"
[17] "summary.POSIXct"
[19] "summary.table"
"summary.aov"
"summary.connection"
"summary.default"
"summary.glm"
"summary.infl"
"summary.lm.null"
"summary.matrix"
"summary.packageStatus"
"summary.POSIXlt"
We can see the difference for this generic in the case of a linear regression,
compared to an analysis of variance, with a small simulated example:
> x <- y <- rnorm(5)
> lm.spray <- lm(y ~ x)
> names(lm.spray)
[1] "coefficients" "residuals"
[4] "rank"
"fitted.values"
[7] "qr"
"df.residual"
[10] "call"
"terms"
> names(summary(lm.spray))
[1] "call"
"terms"
[4] "coefficients" "sigma"
[7] "r.squared"
"adj.r.squared"
[10] "cov.unscaled"
"effects"
"assign"
"xlevels"
"model"
"residuals"
"df"
"fstatistic"
The following table shows some generic functions that do supplementary
analyses from an object resulting from an analysis, the main argument being
60
this latter object, but in some cases a further argument is necessary like for
predict or update.
add1
drop1
step
anova
predict
update
tests successively all the terms that can be added to a model
tests successively all the terms that can be removed from a model
selects a model with AIC (calls add1 and drop1)
computes a table of analysis of variance or deviance for one or several
models
computes the predicted values for new data from a fitted model
re-fits a model with a new formula or new data
There are also various utilities functions that extract information from a
model object or a formula, such as alias which finds the linearly dependent
terms in a linear model specified by a formula.
Finally, there are, of course, graphical functions such as plot which displays various diagnostics, or termplot (see the above example), though this
latter function is not generic but calls predict.
5.4
Packages
The following table lists the standard packages which are distributed with a
base installation of R. Some of them are loaded in memory when R starts; this
can be displayed with the function search:
> search()
[1] ".GlobalEnv"
[3] "package:stats"
[5] "package:grDevices"
[7] "package:datasets"
[9] "package:base"
"package:methods"
"package:graphics"
"package:utils"
"Autoloads"
The other packages may be used after being loaded:
> library(grid)
The list of the functions in a package can be displayed with:
> library(help = grid)
or by browsing the help in html format. The information relative to each
function can be accessed as previously seen (p. 7).
61
Package
Description
base
datasets
grDevices
graphics
grid
methods
base R functions
base R datasets
graphics devices for base and grid graphics
base graphics
grid graphics
definition of methods and classes for R objects and programming tools
regression spline functions and classes
statistical functions
statistical functions using S4 classes
functions to interface R with Tcl/Tk graphical user interface
elements
tools for package development and administration
R utility functions
splines
stats
stats4
tcltk
tools
utils
Many contributed packages add to the list of statistical methods available
in R. They are distributed separately, and must be installed and loaded in
R. A complete list of the contributed packages, with descriptions, is on the
CRAN Web site18 . Several of these packages are recommanded since they cover
statistical methods often used in data analysis. The recommended packages
are often distributed with a base installation of R. They are briefly described
in the following table.
Package
Description
boot
class
cluster
foreign
resampling and bootstraping methods
classification methods
clustering methods
functions for reading data stored in various formats (S3,
Stata, SAS, Minitab, SPSS, Epi Info)
methods for kernel smoothing and density estimation (including bivariate kernels)
Lattice (Trellis) graphics
contains many functions, tools and data sets from the libraries of “Modern Applied Statistics with S” by Venables
& Ripley
generalized additive models
linear and non-linear mixed-effects models
neural networks and multinomial log-linear models
recursive partitioning
spatial analyses (“kriging”, spatial covariance, . . . )
survival analyses
KernSmooth
lattice
MASS
mgcv
nlme
nnet
rpart
spatial
survival
18
http://cran.r-project.org/src/contrib/PACKAGES.html
62
There are two other main repositories of R packages: the Omegahat Project
for Statistical Computing19 which focuses on web-based applications and interfaces between softwares and languages, and the Bioconductor Project 20
specialized in bioinformatic applications (particularly for the analysis of microarray data).
The procedure to install a package depends on the operating system and
whether R was installed from the sources or pre-compiled binaries. In the
latter situation, it is recommended to use the pre-compiled packages available
on CRAN’s site. Under Windows, the binary Rgui.exe has a menu “Packages”
allowing to install packages via internet from the CRAN Web site, or from
zipped files on the local disk.
If R was compiled, a package can be installed from its sources which are
distributed as a ‘.tar.gz’ file. For instance, if we want to install the package gee,
we will first download the file gee 4.13-6.tar.gz (the number 4.13-6 indicates
the version of the package; generally only one version is available on CRAN).
We will then type from the system (and not in R) the command:
R CMD INSTALL gee_4.13-6.tar.gz
There are several useful functions to manage packages such as installed.
packages, CRAN.packages, or download.packages. It is also useful to type
regularly the command:
> update.packages()
which checks the versions of the packages installed against those available
on CRAN (this command can be called from the menu “Packages” under
Windows). The user can then update the packages with more recent versions
than those installed on the computer.
19
20
http://www.omegahat.org/R/
http://www.bioconductor.org/
63
6
Programming with R in pratice
Now that we have done an overview of R’s functionalities, let us return to the
language and programming. We will see a few simple ideas likely to be used
in practice.
6.1
Loops and vectorization
An advantage of R compared to softwares with pull-down menus is the possibility to program simply a series of analyses which will be executed successively.
This is common to any computer language, but R has some particular features
which make programming easier for non-specialists.
Like other languages, R has some control structures which are not dissimilar to those of the C language. Suppose we have a vector x, and for each
element of x with the value b, we want to give the value 0 to another variable
y, otherwise 1. We first create a vector y of the same length than x:
y <- numeric(length(x))
for (i in 1:length(x)) if (x[i] == b) y[i] <- 0 else y[i] <- 1
Several instructions can be executed if they are placed within braces:
for (i in 1:length(x)) {
y[i] <- 0
...
}
if (x[i] == b) {
y[i] <- 0
...
}
Another possible situation is to execute an instruction as long as a condition is true:
while (myfun > minimum) {
...
}
However, loops and control structures can be avoided in most situations
thanks to a feature of R: vectorization. Vectorization makes loops implicit in
expression, and we have seen many cases. Let us consider the addition of two
vectors:
64
> z <- x + y
This addition could be written with a loop, as this is done in most languages:
> z <- numeric(length(x))
> for (i in 1:length(z)) z[i] <- x[i] + y[i]
In this case, it is necessary to create the vector z beforehand because of
the use of the indexing system. We realize that this explicit loop will work
only if x and y are of the same length: it must be changed if this is not true,
whereas the first expression will work in all situations.
The conditional executions (if ... else) can be avoided with the use
of the logical indexing; coming back to the above example:
> y[x == b] <- 0
> y[x != b] <- 1
In addition to being simpler, vectorized expressions are computationally
more efficient, particularly with large quantities of data.
There are also several functions of the type ‘apply’ which avoids writing
loops. apply acts on the rows and/or columns of a matrix, its syntax is
apply(X, MARGIN, FUN, ...), where X is a matrix, MARGIN indicates whether
to consider the rows (1), the columns (2), or both (c(1, 2)), FUN is a function
(or an operator, but in this case it must be specified within brackets) to apply,
and ... are possible optional arguments for FUN. A simple example follows.
>
>
>
>
x <- rnorm(10, -5, 0.1)
y <- rnorm(10, 5, 2)
X <- cbind(x, y) # the columns of X keep the names "x" and "y"
apply(X, 2, mean)
x
y
-4.975132 4.932979
> apply(X, 2, sd)
x
y
0.0755153 2.1388071
lapply() acts on a list: its syntax is similar to apply and it returns a list.
> forms <- list(y ~ x, y ~ poly(x, 2))
> lapply(forms, lm)
[[1]]
Call:
FUN(formula = X[[1]])
Coefficients:
65
(Intercept)
31.683
x
5.377
[[2]]
Call:
FUN(formula = X[[2]])
Coefficients:
(Intercept) poly(x, 2)1
4.9330
1.2181
poly(x, 2)2
-0.6037
sapply() is a flexible variant of lapply() which can take a vector or a
matrix as main argument, and returns its results in a more user-friendly form,
generally as a table.
6.2
Writing a program in R
Typically, an R program is written in a file saved in ASCII format and named
with the extension ‘.R’. A typical situation where a program is useful is when
one wants to do the same tasks several times. In our first example, we want
to do the same plot for three different species of birds, the data being in three
distinct files. We will proceed step by step, and see different ways to program
this very simple problem.
First, let us make our program in the most intuitive way by executing successively the needed commands, taking care to partition the graphical device
beforehand.
layout(matrix(1:3, 3, 1))
data <- read.table("Swal.dat")
plot(data$V1, data$V2, type="l")
title("swallow")
data <- read.table("Wren.dat")
plot(data$V1, data$V2, type="l")
title("wren")
data <- read.table("Dunn.dat")
plot(data$V1, data$V2, type="l")
title("dunnock")
# partition the graphics
# read the data
# add a title
The character ‘#’ is used to add comments in a program: R then goes to
the next line.
The problem of this first program is that it may become quite long if we
want to add other species. Moreover, some commands are executed several
times, thus they can be grouped together and executed after changing some
arguments. The strategy used here is to put these arguments in vectors of
mode character, and then use the indexing to access these different values.
66
layout(matrix(1:3, 3, 1))
# partition the graphics
species <- c("swallow", "wren", "dunnock")
file <- c("Swal.dat" , "Wren.dat", "Dunn.dat")
for(i in 1:length(species)) {
data <- read.table(file[i])
# read the data
plot(data$V1, data$V2, type="l")
title(species[i])
# add a title
}
Note that there are no double quotes around file[i] in read.table()
since this argument is of mode character.
Our program is now more compact. It is easier to add other species since
the vectors containing the species and file names are at the beginning of the
program.
The above programs will work correctly if the data files ‘.dat’ are located
in the working directory of R, otherwise the user must either change the working directory, or specifiy the path in the program (for example: file <"/home/paradis/data/Swal.dat"). If the program is written in the file Mybirds.R, it will be called by typing:
> source("Mybirds.R")
Like for any input from a file, it is necessary to give the path to access the
file if it is not in the working directory.
6.3
Writing your own functions
We have seen that most of R’s work is done with functions which arguments
are given within parentheses. Users can write their own functions, and these
will have exactly the same properties than other functions in R.
Writing your own functions allows an efficient, flexible, and rational use of
R. Let us come back to our example of reading some data followed by plotting
a graph. If we want to do this operation in different situations, it may be a
good idea to write a function:
myfun <- function(S, F)
{
data <- read.table(F)
plot(data$V1, data$V2, type="l")
title(S)
}
To be executed, this function must be loaded in memory, and this can be
done in several ways. The lines of the function can be typed directly on the
keyboard, like any other command, or copied and pasted from an editor. If
the function has been saved in a text file, it can be loaded with source()
67
like another program. If the user wants some functions to be loaded each
time when R starts, they can be saved in a workspace .RData which will be
loaded in memory if it is in the working directory. Another possibility is to
configure the file ‘.Rprofile’ or ‘Rprofile’ (see ?Startup for details). Finally,
it is possible to create a package, but this will not be discussed here (see the
manual “Writing R Extensions”).
Once the function is loaded, we will be able with a single command to read
the data and plot the graph, for instance with myfun("swallow", "Swal.dat").
Thus, we have now a third version of our program:
layout(matrix(1:3, 3, 1))
myfun("swallow", "Swal.dat")
myfun("wren", "Wrenn.dat")
myfun("dunnock", "Dunn.dat")
We may also use sapply() leading to a fourth version of our program:
layout(matrix(1:3, 3, 1))
species <- c("swallow", "wren", "dunnock")
file <- c("Swal.dat" , "Wren.dat", "Dunn.dat")
sapply(species, myfun, file)
In R, it is not necessary to declare the variables used within a function.
When a function is executed, R uses a rule called lexical scoping to decide
whether an object is local to the function, or global. To understand this
mechanism, let us consider the very simple function below:
> foo <- function() print(x)
> x <- 1
> foo()
[1] 1
The name x is not used to create an object within foo(), so R will seek
in the enclosing environment if there is an object called x, and will print its
value (otherwise, a message error is displayed, and the execution is halted).
If x is used as the name of an object within our function, the value of x in
the global environment is not used.
> x <- 1
> foo2 <- function() { x <- 2; print(x) }
> foo2()
[1] 2
> x
[1] 1
This time print() uses the object x that is defined within its environment,
that is the environment of foo2.
68
The word “enclosing” above is important. In our two example functions,
there are two environments: the global one, and the one of the function foo
or foo2. If there are three or more nested environments, the search for the
objects is made progressively from a given environment to the enclosing one,
and so on, up to the global one.
There are two ways to specify arguments to a function: by their positions
or by their names (also called tagged arguments). For example, let us consider
a function with three arguments:
foo <- function(arg1, arg2, arg3) {...}
foo() can be executed without using the names arg1, . . . , if the corresponding objects are placed in the correct position, for instance: foo(x, y,
z). However, the position has no importance if the names of the arguments
are used, e.g. foo(arg3 = z, arg2 = y, arg1 = x). Another feature of
R’s functions is the possibility to use default values in their definition. For
instance:
foo <- function(arg1, arg2 = 5, arg3 = FALSE) {...}
The commands foo(x), foo(x, 5, FALSE), and foo(x, arg3 = FALSE) will
have exactly the same result. The use of default values in a function definition
is very useful, particularly when used with tagged arguments (i.e. to change
only one default value such as foo(x, arg3 = TRUE).
To conclude this section, let us see another example which is not purely
statistical, but it illustrates the flexibility of R. Consider we wish to study the
behaviour of a non-linear model: Ricker’s model defined by:
Nt+1
Nt
= Nt exp r 1 −
K
This model is widely used in population dynamics, particularly of fish. We
want, using a function, to simulate this model with respect to the growth rate
r and the initial number in the population N 0 (the carrying capacity K is
often taken equal to 1 and this value will be taken as default); the results will
be displayed as a plot of numbers with respect to time. We will add an option
to allow the user to display only the numbers in the last few time steps (by
default all results will be plotted). The function below can do this numerical
analysis of Ricker’s model.
ricker <- function(nzero, r, K=1, time=100, from=0, to=time)
{
N <- numeric(time+1)
N[1] <- nzero
for (i in 1:time) N[i+1] <- N[i]*exp(r*(1 - N[i]/K))
Time <- 0:time
plot(Time, N, type="l", xlim=c(from, to))
}
69
Try it yourself with:
>
>
>
>
layout(matrix(1:3, 3, 1))
ricker(0.1, 1); title("r = 1")
ricker(0.1, 2); title("r = 2")
ricker(0.1, 3); title("r = 3")
70
7
Literature on R
Manuals. Several manuals are distributed with R in R HOME/doc/manual/:
• An Introduction to R [R-intro.pdf],
• R Installation and Administration [R-admin.pdf],
• R Data Import/Export [R-data.pdf],
• Writing R Extensions [R-exts.pdf],
• R Language Definition [R-lang.pdf].
The files may be in different formats (pdf, html, texi, . . . ) depending on
the type of installation.
FAQ. R is also distributed with an FAQ (Frequently Asked Questions) localized in the directory R HOME/doc/html/. A version of the R-FAQ is
regularly updated on CRAN’s Web site:
http://cran.r-project.org/doc/FAQ/R-FAQ.html
On-line resources. The CRAN Web site hosts several documents, bibliographic resources, and links to other sites. There are also a list of publications (books and articles) about R or statistical methods 21 and some
documents and tutorials written by R’s users 22 .
Mailing lists. There are four discussion lists on R; to subscribe, send a message, or read the archives see: http://www.R-project.org/mail.html.
The general discussion list ‘r-help’ is an interesting source of information
for the users of R (the three other lists are dedicated to annoucements
of new versions, and for developers). Many users have sent to ‘r-help’
functions or programs which can be found in the archives. If a problem
is encountered with R, it is thus important to proceed in the following
order before sending a message to ‘r-help’:
1. read carefully the on-line help (possibly using the search engine);
2. read the R-FAQ;
3. search the archives of ‘r-help’ at the above address, or by using one
of the search engines developed on some Web sites 23 ;
4. read the “posting guide”24 before sending your question(s).
21
http://www.R-project.org/doc/bib/R-publications.html
http://cran.r-project.org/other-docs.html
23
The addresses of these sites are listed at http://cran.r-project.org/search.html
24
http://www.r-project.org/posting-guide.html
22
71
R News. The electronic journal R News aims to fill the gap between the
electronic discussion lists and traditional scientific publications. The
first issue was published on January 2001 25 .
Citing R in a publication. Finally, if you mention R in a publication, you
must cite the following reference:
R Development Core Team (2005). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL:
http://www.R-project.org.
25
http://cran.r-project.org/doc/Rnews/
72
Download